Risk factors associated with poor pain outcomes following primary knee replacement surgery: Analysis of data from the clinical practice research datalink, hospital episode statistics and patient reported outcomes as part of the STAR research programme.

OBJECTIVE: Identify risk factors for poor pain outcomes six months after primary knee replacement surgery.
METHODS: Observational cohort study on patients receiving primary knee replacement from the UK Clinical Practice Research Datalink, Hospital Episode Statistics and Patient Reported Outcomes. A wide range of variables routinely collected in primary and secondary care were identified as potential predictors of worsening or only minor improvement in pain, based on the Oxford Knee Score pain subscale. Results are presented as relative risk ratios and adjusted risk differences (ARD) by fitting a generalized linear model with a binomial error structure and log link function.
RESULTS: Information was available for 4,750 patients from 2009 to 2016, with a mean age of 69, of whom 56.1% were female. 10.4% of patients had poor pain outcomes. The strongest effects were seen for pre-operative factors: mild knee pain symptoms at the time of surgery (ARD 18.2% (95% Confidence Interval 13.6, 22.8), smoking 12.0% (95% CI:7.3, 16.6), living in the most deprived areas 5.6% (95% CI:2.3, 9.0) and obesity class II 6.3% (95% CI:3.0, 9.7). Important risk factors with more moderate effects included a history of previous knee arthroscopy surgery 4.6% (95% CI:2.5, 6.6), and use of opioids 3.4% (95% CI:1.4, 5.3) within three months after surgery. Those patients with worsening pain state change had more complications by 3 months (11.8% among those in a worse pain state vs. 2.7% with the same pain state).
CONCLUSIONS: We quantified the relative importance of individual risk factors including mild pre-operative pain, smoking, deprivation, obesity and opioid use in terms of the absolute proportions of patients achieving poor pain outcomes. These findings will support development of interventions to reduce the numbers of patients who have poor pain outcomes.

Introduction

Knee replacement surgery may be offered to patients with knee osteoarthritis who have not responded to conservative treatment [1]. Over 100,000 knee replacement operations are carried out each year in the UK for osteoarthritis and other surgical indications [2]. Many patients can expect to achieve reductions in knee pain and improvements in functional outcomes following surgery [3]. The percentage who experience ongoing chronic knee pain post-surgery is variable [4], with up to 20% experiencing knee pain that impacts their quality of life after 3 months post-op [5]. Patients who experience this kind of pain after surgery have not received the expected benefit and for some their pain is worse than it was before the operation [6, 7].

It is important to identify which patients are at greatest risk of similar or worse pain after knee replacement. When healthcare professionals and patients are making decisions about treatment options, knowledge of the chances of benefit and the risk of harms, or of no benefit, is key to informed choice. Although previous research has explored predictors of outcomes of knee replacement [8], most studies have focused on total scores encompassing several domains (e.g. pain, stiffness and function) and fewer studies have focussed solely on pain status [7]. Most existing studies tend to use continuous patient-reported outcome scores, but as the majority of patients achieve good outcomes, they can only help us identify predictors that differentiate between patients that achieve a ‘really good outcome’ versus a ‘good outcome’ [9, 10]. Instead, defining pain outcome by focussing on those patients whose symptoms have had no clinically meaningful change or which worsened after surgery, would allow identification of poor outcomes [11-14]. The predictive ability of previous studies is poor, which means that they are unable to explain variation in patient-reported pain after knee replacement [7, 10]. Most research has explored pre-operative risk factors, but little research has sought to understand a wider range of pain determinants that occur post-operatively. These are important because the time after surgery may present an ideal opportunity for targeted intervention to prevent the persistence or worsening of pain.

The aim of this study is to identify pre- and post-operative risk factors for whether or not a patient has a poor pain outcome after knee replacement surgery, by analysing a wide range of potential factors from the UK Clinical Practice Research Datalink (CPRD) primary care GOLD database linked to English Hospital Episode Statistics (HES) hospital admissions and to Patient Reported Outcomes Measures (PROMs) data.

Methods

Study design

Retrospective observational study using anonymised linked data from CPRD, HES and PROMs.

Data source

CPRD GOLD contains anonymised individual patient data from electronic primary healthcare records from practices across the United Kingdom [15]. CPRD is one of the largest databases of longitudinal primary care medical records in the world with coverage of 674 general practices in the UK with 11.3 million patients, of which 4.4 million patients are active [15]. Primary care records from CPRD were linked to secondary care admission records from HES Admitted Patient Care data and to Office for National Statistics (ONS) mortality data. From 1st April 2009, HES provides PROMs data before and six months following knee replacements. Linkage of CPRD-HES-ONS-PROMS data is done by NHS Digital as a ‘trusted third party’.

Sample

We included all patients receiving a primary total or uni-compartmental knee replacement between 2009 and 2016. Inclusion in the analysis was limited to those patients with HES linked data (England only) who completed both the pre- and six-month post-operative Oxford Knee Score pain subscale (see Flow Diagram Fig 1).

Fig 1

Patient flow diagram.

TKR/UKR, total and uni-compartmental knee replacement; CPRD, Clinical Practice Research Datalink GOLD; HES, English Hospital Episode Statistics; PROMs, Patient Reported Outcome Measures; OKS, Oxford Knee Score; Underweight BMI, Body Mass Index under 18.5 Kg/m2.

Main outcome measure

The Oxford Knee score (OKS) [16] is collected as part of the national PROMs programme and is a measure of patient-reported pain and function. Each of the 12 questions is scored between 0 (meaning worst symptoms) and 4 (least troublesome symptoms). Pain- and function-related subscales within the OKS have been identified and validated [17]. An OKS pain subscale (OKS-PS) summary score can be calculated, ranging from 0 (most pain) to 28 (least pain), by summing the responses of the seven OKS-PS items.

The Treatment Effect (TE = (pre-treatment score − post treatment score)/pre-treatment score) [12, 18] was calculated for each patient using the OKS-PS score (normalized to a score from 0 (least pain) to 100 (worst pain)). A TE of 1 (best score) corresponds to a patient without pain after treatment, a TE of 0 to no improvement and no deterioration, a negative TE to more pain at follow up. TE ≤0.2 is used to classify whether or not a patient responds to surgery in respect of their knee pain, and has previously been validated against the OARSI-OMERACT criteria [14] to identify responders to surgery [12].

Predictor variables

In discussion with clinicians we sought to identify variables within this routinely collected dataset which may provide a wide range of potential predictors for the pain outcome.

Pre-operative predictors

To measure socioeconomic deprivation, we used the index of multiple deprivation (IMD) score. This is a relative measure of deprivation for small areas—termed lower super output areas (LSOAs)–which are defined as geographical areas of a similar population size, with an average of 1,500 residents [19]. The IMD comprises seven measures of deprivation: income deprivation; employment deprivation; education, skills and training deprivation; health deprivation and disability; crime; barriers to housing and services; and living environment deprivation. We used the IMD rank for a patient’s LSOA and categorised patients into quintiles based upon the national ranking of local areas, with quintile 1 being the least deprived group and quintile 5 being the most deprived group (i.e. reordered to aid reporting). As a measure of comorbidity we used the Royal College of Surgeons’ (RCS) Charlson Score, which is calculated based on the presence of several chronic conditions identified using ICD-10 codes at the time of knee replacement surgery admission and all admissions in the preceding 5 years [20].

Patient case-mix factors included: pre-operative OKS pain score, age at surgery, Body Mass Index (BMI), smoking, alcohol consumption, gender, index of multiple deprivation (IMD) score, Charlson Comorbidity Index (five years prior to surgery). Previous medication use included steroids (non-glucocorticosteroids (non-GCS), Steroid (any type of injection), oral), non-steroidal anti-inflammatory drugs (NSAIDS), opioids, antibiotics, and antidepressants. We identified whether the primary procedure was a total or uni-compartmental knee replacement. We classified patients according to whether or not they had a knee arthroscopy prior to surgery.

Post-operative predictors

Length of stay (LOS) at hospital was calculated as the number of days between the hospital admission and discharge date. We identified medical complications as one or more events happening 3 months after the operation from the following list: stroke (excluding mini stroke), respiratory infection, acute myocardial infarction, pulmonary embolism/deep vein thrombosis, urinary tract infection, wound disruption, surgical site infection, fracture after implant, complication of prosthesis, neurovascular injury, acute renal failure and blood transfusion [21].

Re-operations include stiffness requiring manipulation under anaesthetic (MUA), arthroscopic surgery, debridement for infection and operations for wound problems. Re-operations included open operations (such as debridements for infection or ligament repairs), arthroscopic operations (excision of loose bodies or menisci in uni-compartmental knee replacement (UKR), washouts/debridements for infection), and closed operations. We also evaluated the rate of revision by three months after the surgery.

We identified medications prescribed (including opioids, NSAIDS, and antibiotics) pre- and post-surgery and calculated the total number of general practice visits between surgery and 3 months post-surgery. We have assessed for evidence of collinearity using variance inflation factors and there was no evidence of multicollinearity.

Statistical analysis

To describe their change in pain state before and 6 months after surgery, patients are categorised into pain groups, and descriptive statistics (without statistical testing) used to describe the number (percentage) of patients that move between different pain states before and after surgery. We then describe the characteristics of patients who are most likely to be moving to a worse pain state post-operatively.

Logistic regression modeling was used to describe the association of predictor variables with the outcome of interest (responder to surgery according to TE pain score). As our dataset is large, we selected the lowest category for each variable as the reference category. Composite variables were used if individual characteristics were rare. Results of the regression model are presented as relative risk ratios by fitting a generalized linear model with a binomial error structure and a log link function (log-logistic model). Results are further presented as adjusted risk differences estimated from marginal effects from the logistic regression model [22]. Fractional polynomial regression was used to assess evidence of linearity of continuous predictors with the outcome. As there was evidence of non-linearity for the pre-operative OKS pain score and BMI, these variables were categorized. We excluded nine underweight BMI patients, as there were too few patients in this category, leading to a small cell problem in the multivariable regression model, and it would be inappropriate to combine underweight and normal BMI categories together. Multiple imputation by chained equations was used to account for the cumulative effect of missing data in several of the variables [23]. Forty imputed datasets were generated using all potential factors (including the outcome) and estimated parameters were combined using Rubin’s rules. The C-statistic was used to describe the discriminatory ability of variables in the final model. We examine the strength of associations and not arbitrary measures of statistical significance with cut offs of for example p<0.05 or the related concept of whether the confidence interval includes the null value.

Analyses were conducted using Stata version 15.1 statistical software (StataCorp, College Station, Texas). We followed the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guideline to report our study [24].

Ethics approval

The CPRD Group has obtained ethical approval from a National Research Ethics Service Committee (NRES) for all purely observational research using anonymised CPRD data; namely, studies which do not include patient involvement. The study has been approved by ISAC (Independent Scientific Advisory Committee) for MHRA Database Research) (protocol number 11_050AMnA2RA2).

Results

Information was available for 4,750 patients over the time period from 2009 to 2016 (Fig 1), with a mean age of 69 (SD 9), of whom 56.1% were female. Only four of the pre-operative variables had missing data: BMI (352, 7.4%), smoking (81, 1.7%), alcohol consumption (816, 17.2%) and IMD (5, 0.1%). Multiple imputation was used to account for missing data in the analysis. Assumptions of the imputation model were assessed and described in the (S1 Table).

Within the dataset, 10.4% of patients experienced a poor pain outcome following surgery, such that their symptoms of pain did not have a relative improvement of at least 20% or worsened by six months after surgery. Fig 2 shows the distribution of the treatment effects score; although the majority of patients had a good pain outcome (blue bars), there is a minority that have poor pain outcomes (red bars).

Fig 2

Distribution of the treatment effect score for patients who did, and did not, respond to surgery.

Red = poor pain outcome, Blue = good pain outcome.

Pre-operative predictors

The effect of pre-operative OKS-PS had a non-linear association with pain outcome (Fig 3). Patients with the mildest pain symptoms (OKS-PS 17–28) had a 3.9 times increased risk of a poor pain outcome compared to those with OKS-PS 8–10. A ‘U-shaped’ effect was observed for patient age at surgery, where those aged between 60–69 and 70–79 had a reduced risk of poor pain outcome compared with the youngest age group. The oldest age group was similar to the youngest. Compared with those of normal BMI, being overweight and obese increased risk of poor pain response to surgery. The highest proportion of patients with poor pain outcomes were in the group of current smokers, males, people living in the most deprived areas and those with inflammatory arthritis (Table 1). For the following variables there is weak evidence of an association. An effect of Charlson co-morbidities was only seen for those with four or more co-morbidities getting worse outcomes. Having a uni-compartmental rather than total knee replacement was associated with a better pain response, but this was attenuated when post-operative factors were included in the model. Patients who had previously had knee arthroscopy prior to their primary knee replacement were more likely to have poor pain outcomes. Patients prescribed full opioids and antidepressants were more likely to have a poor pain response.

Fig 3

Forest plot of predictors of poor pain outcomes.

Table 1

Descriptive statistics describing the total number of patients with each potential risk factor, and the proportion of patients with a poor pain outcome, according to whether or not they have the factor.

	Total	Proportion of patients with poor pain response with and without each risk factor
		Without	With
PRE-OPERATIVE
Pre-op OKS pain score
<5	562 (11.8%)		59 (10.5%)
5 to 7	967 (20.4%)		97 (10.0%)
8 to 10	1123 (23.6%)		77 (6.9%)
11 to 13	1039 (21.9%)		113 (10.9%)
14 to 16	658 (13.9%)		68 (10.3%)
17 to 28	401 (8.4%)		80 (20.0%)
Age (years)
<60	703 (14.8%)		110 (15.7%)
60 to 69	1727 (36.4%)		174 (10.1%)
70 to 79	1741 (36.7%)		155 (8.9%)
80+	579 (12.2%)		55 (9.5%)
BMI
Normal	650 (14.8%)		42 (6.5%)
Overweight	1692 (38.5%)		175 (10.3%)
Obese class I	1236 (28.1%)		146 (11.8%)
Obese class II	589 (13.4%)		76 (12.9%)
Obese class III	231 (5.3%)		25 (10.8%)
Smoking
Ex	1776 (38.0%)		188 (10.6%)
No	2598 (55.6%)		231 (8.9%)
Yes	295 (6.3%)		71 (24.1%)
Drinking
Ex	117 (3.0%)		17 (14.5%)
No	647 (16.5%)		75 (11.6%)
Yes	3170 (80.6%)		309 (9.8%)
Gender
Female	2664 (56.1%)		246 (9.2%)
Male	2086 (43.9%)		248 (11.9%)
IMD deprivation score (quintiles)
1—Least deprived	1185 (25.0%)		98 (8.3%)
2	1225 (25.8%)		117 (9.6%)
3	1055 (22.2%)		101 (9.6%)
4	789 (16.6%)		94 (11.9%)
5—Most deprived	491 (10.4%)		84 (17.1%)
Charlson Comorbidity (5-years prior)
None	3341 (70.3%)		329 (9.9%)
1	385 (8.1%)		40 (10.4%)
2	593 (12.5%)		63 (10.6%)
3	189 (4.0%)		26 (13.8%)
4+	242 (5.1%)		36 (14.9%)
Comorbidities
Hypertension	1944 (40.9%)	279 (9.9%)	215 (11.1%)
Hyperlipidaemia	811 (17.1%)	402 (10.2%)	92 (11.3%)
Ischaemic heart disease (IHD)	381 (8.0%)	446 (10.2%)	48 (12.6%)
Cardiovascular disease (CVD)	142 (3.0%)	477 (10.4%)	17 (12.0%)
Chronic obstructive pulmonary disease (COPD)	146 (3.0%)	470 (10.2%)	24 (16.4%)
Renal failure	625 (13.2%)	426 (10.3%)	68 (10.9%)
Cancer	538 (11.3%)	437 (10.4%)	57 (10.6%)
Rheumatoid arthritis	125 (2.6%)	485 (10.5%)	9 (7.2%)
Lupus	8 (0.2%)	493 (10.4%)	1 (12.5%)
Inflammatory arthritis	5 (0.1%)	493 (10.4%)	1 (20.0%)
Ankylosing Spondylitis	23 (0.5%)	492 (10.4%)	2 (8.7%)
Diabetes	573 (12.1%)	415 (9.9%)	79 (13.8%)
Knee replacement
Total	4212 (88.7%)		449 (10.7%)
Uni-compartmental	538 (11.3%)		45 (8.4%)
Knee arthroscopy	1416 (29.8%)	288 (8.6%)	206 (14.6%)
Medications
Steroids non-GCS	8 (0.2%)	493 (10.4%)	1 (12.5%)
Steroids GCS injections	1829 (38.5%)	292 (10.0%)	202 (11.0%)
Steroids oral	1136 (23.9%)	366 (10.1%)	128 (11.3%)
Prednisolone	1119 (23.6%)	368 (10.1%)	126 (11.3%)
NSAIDs	4226 (89.0%)	50 (9.5%)	444 (10.5%)
Opioids (full)	2022 (42.6%)	226 (8.3%)	268 (13.3%)
Opioids (partial)	3517 (74.0%)	105 (8.5%)	389 (11.1%)
Antibiotics	4313 (90.8%)	29 (6.6%)	465 (10.8%)
Anticonvulsants (gabapentin, pregabalin)	410 (8.6%)	430 (9.9%)	64 (15.6%)
Paracetamol	3910 (82.3%)	78 (9.3%)	416 (10.6%)
Antidepressants (SSRI, TCA)	1949 (41.0%)	245 (8.8%)	249 (12.8%)
POST-OPERATIVE
Complication (3-months)	203 (4.3%)	463 (10.2%)	31 (15.3%)
Readmission (3-months)	568 (12.0%)	396 (9.5%)	98 (17.3%)
Re-operation (3-months)	141 (3.0%)	470 (10.2%)	24 (17.0%)
Revision (3-months)	8 (0.2%)	491 (10.4%)	3 (37.5%)
Manipulation under anaesthetic (3-months)	42 (0.9%)	481 (10.2%)	13 (31.0%)
Irrigation / Debridement (3-months)	27 (0.6%)	488 (10.3%)	6 (22.2%)
Length of stay (primary)
< 2-days	56 (1.2%)		5 (8.9%)
2 to 4 days	1808 (38.1%)		167 (9.2%)
4 to 6 days	1853 (39.0%)		200 (10.8%)
6 to 10 days	799 (16.8%)		87 (10.9%)
>10 days	234 (4.9%)		35 (15.0%)
Number of GP consultations (3-months)
None	306 (6.4%)		24 (7.8%)
1 to 4	2680 (56.4%)		235 (8.8%)
5 to 9	1395 (29.4%)		185 (13.3%)
10+	369 (7.8%)		50 (13.6%)
Medication use (3-months)
NSAIDS	1695 (35.7%)	304 (10.0%)	190 (11.2%)
Opioids (full)	1681 (35.4%)	253 (8.2%)	241 (14.3%)
Opioids (partial)	1136 (23.9%)	368 (10.2%)	126 (11.1%)
Antibiotics	1117 (23.5%)	340 (9.4%)	154 (13.8%)
Anticonvulsants (gabapentin, pregabalin)	191 (4.0%)	471 (10.3%)	23 (12.0%)
Paracetamol	2241 (47.2%)	256 (10.2%)	238 (10.6%)
Antidepressants	738 (15.5%)	389 (9.7%)	105 (14.2%)

Post-operative predictors

Medical complications occurring within three months of surgery were rare (Table 1) with an overall complication rate of 4.3%. The overall rate of re-operation was 3.0% and only eight (0.2%) patients were revised within three months. Re-admission to hospital for any reason after surgery was more common at 12.0%.

Re-admission to hospital, revision surgery, and manipulation under anaesthetic within three months of the operation were all associated with poor pain response to surgery at six months (Fig 3). In respect of medication use post operatively, patients prescribed opioids and antibiotics had a stronger association with poor pain response.

Absolute risk differences

Table 1 presents descriptive statistics showing the observed proportions of poor outcomes for each category of the predictor variables. Fig 4 shows the adjusted risk differences. For the pre-operative OKS-PS, this can be interpreted as those having a score of 17–28 with a poor pain outcome 18.2 percentage points more often (95% Confidence Interval 13.6, 22.8) than with a pre-operative score of 8–10, on average. The other pre-operative variables with some absolute adjusted risk differences were smoking 12.0 percentage points (95% CI: 7.3, 16.6), obesity class II 6.3 percentage points (95% CI: 3.0, 9.7), and living in the most deprived areas 5.6 percentage points (95% CI: 2.3, 9.0). For the post-operative risk factors, revision surgery 15.8 percentage points (95% CI: -10.2, 41.7) and manipulation under anaesthetic 9.7 percentage points (95% CI: -1.1, 20.5) conferred some absolute adjusted risk of poor pain outcome, although these were not statistically significant. Important risk factors with more moderate effects included a history of previous knee arthroscopy surgery 4.6 percentage points (95% CI:2.5, 6.6), and use of opioids 3.4 percentage points (95% CI:1.4, 5.3) within three months after surgery.

Fig 4

Adjusted risk differences for predictors of poor pain outcomes.

The discriminatory ability of variables in the final model was modest (c-statistic 0.72, 95% Confidence Interval 0.70, 0.75) (S1 Fig).

Pain state change

Table 2 characterises patients’ change in pain status before and at 6 months after surgery. Patients with the mildest pre-operative pain symptoms (OKS-PS 17–28) were most likely to not improve and move to a worse pain state following surgery, where 20% of patients had a poor pain outcome compared to around 10% in the other pre-operative pain states. To further understand why patients in this mild pre-operative pain state (OKS-PS 17–28) had worse pain outcomes, we describe the pre-operative and post-operative characteristics of those with worsening pain state change. These patients had more complications by 3 months (11.8% among those in a worse pain state vs. 2.7% with the same pain state); readmission (26.5% vs. 8.2%); and re-operation (5.9% vs. 3.0%). There were some differences in length of stay, where those who moved to a worse pain state had much shorter length of stay, particularly < 2-days (23.5% vs. 1.9%), with more pain medication use pre-operatively [(NSAIDS (94.1% vs. 85.0%), opioids (full) (35.3% vs. 25.6%); opioids (partial) (79.4% vs. 57.0%) and antibiotics (32.4% vs. 18.3%)]; and at 3 months post-operatively [(opioids (full) (50.0% vs. 22.1%), opioids (partial) (26.5% vs. 18.0%) and antibiotics (32.4% vs. 18.3%), with less marked difference for other medicines]. In terms of characteristics, those who moved to a worse pain state were more likely to be obese and have more pre-operative comorbidities (one or more comorbidities 41.2% vs. 25.95%), particularly diabetes (23.5% vs 9.8%).

Table 2

Pain state change between pre-operative and 6-month post-operative assessments.

	6-month post-op OKS pain score
Pre-op OKS pain score	<4	4 to 6	7 to 9	10 to 12	13 to 16	17 to 28	Poor Pain outcome
<4	26 (4.6%)	36 (6.4%)	40 (7.1%)	54 (9.6%)	68 (12.1%)	338 (60.1%)	59 (10.5%)
4 to 6	16 (1.7%)	31 (3.2%)	52 (5.4%)	83 (8.6%)	105 (10.9%)	680 (70.3%)	97 (10.0%)
7 to 9	7 (0.6%)	14 (1.3%)	22 (2.0%)	60 (5.3%)	91 (8.1%)	929 (82.7%)	77 (6.9%)
10 to 12	4 (0.4%)	13 (1.3%)	19 (1.8%)	40 (3.9%)	79 (7.6%)	884 (85.1%)	113 (10.9%)
13 to 16	1 (0.2%)	1 (0.2%)	7 (1.1%)	17 (2.6%)	27 (4.1%)	605 (92.0%)	68 (10.3%)
17 to 28		5 (1.3%)		7 (1.8%)	22 (5.5%)	367 (91.5%)	80 (20.0%)

Discussion

Main findings

We have identified a number of risk factors that are associated with an increased risk of poor pain outcome. The strongest pre-operative risk factors were: having only mild knee pain symptoms at the time of surgery, being a current smoker, obesity, and living in the most deprived areas. Opioid and antidepressant medication use were also associated with worse pain outcomes. The strongest post-operative factors were revision surgery and manipulation under anaesthetic within three months after the operation. We identified a range of other important risk factors with more moderate effects in terms of absolute risk differences in pain outcome, including a history of previous knee arthroscopy, and use of opioids within the three months after surgery, in addition to a number of other risk factors. Those with the least pre-operative pain were more likely to move to a worse post-operative pain state and were most likely to take pain relieving medicines both pre- and post-surgery, including opioids.

Strengths and limitations

Strengths of this study include the use of a large national linked dataset containing a wide range of clinical information from both primary and secondary care, and in the time periods both before and after surgery. The CPRD-HES linked data have previously been shown to be representative of the wider population in respect of patient demographic characteristics [15, 25]. This sample has also been compared with the mandatory National Joint Registry (NJR) in respect to knee replacement patient profile, for NJR mean age 68.9 (SD 9.6), 56.6% female. In our CPRD sample, the mean age was 69 (SD 9), of whom 56.1% were female. However, within this routine dataset, we do not have information on whether a patient received a unilateral or bilateral knee replacement and hence we are unable to exclude bilateral procedures from the dataset. Another limitation is that we are making an assumption that risk factors of poor pain outcomes, are the same for patients receiving uni-compartmental and total knee replacement. Testing for this would require test for interaction, with all other risk factors in the model, but such multiple testing could lead to type 1 errors and are in any case very low powered. Medical and surgical complications were considered as separate predictors a priori and others may have defined complications differently from this study. PROMs data provided a robust measure of patient-reported pain symptoms. A limitation is that national linked PROMs data has considerable levels of missing data (60.4%) in the six-month post-operative questionnaire. As the six-month data comprises our study’s outcome variable, we only included patients with both pre- and six-month post-operative OKS-PS. There were very little missing data in the wide range of predictor variables included in the study, with the exception of BMI, smoking and drinking, and these variables are widely known to have missing data in CPRD. To account for potential bias, we imputed these variables using multiple imputation. A limitation is the lack of any data on coping, beliefs and expectations of outcome, and any basic mental health data, in particular of depression. The time window we used to define post-operative risk factors of up to three months after surgery was identified on the basis of agreed definitions of chronic post-surgical pain that develops or increases in intensity after a surgical procedure [26] (defined as pain ‘present for at least three months’ [27]), and which may provide a useful window in which clinicians could identify patients and intervene to prevent the development of persistent and potentially intractable pain.

What is already known

Patients with worse pre-operative pain achieve greater change in symptoms (journey), whilst those with mild pre-operative pain have less change but retain the greatest post-operative level of pain and mildest symptoms (destination) [10, 28]. Pre-operative factors age [29], gender [30], obesity [31], social deprivation [7], co-morbidity [32] and smoking [33] are known to influence the surgical outcome, as does having a uni-compartmental knee replacement [34] and the influence of previous non-replacement knee surgery such as knee arthroscopy [35]. This study’s unique contribution is our focus on knee pain as the outcome, and our work to identify patients who do not respond to surgery expressed in terms of adjusted absolute risk. Some other studies have looked at composite outcomes combining symptoms of pain, stiffness and function. Predictors of pain outcomes are not necessarily the same as functional outcomes, and likewise a patient may have improvement in pain symptoms, but not in function [7].

Fewer studies have tried to identify post-operative risk factors [36], and a strength of our study is the focus on primary care risk factors and medication use that is not usually available in other routine datasets. The operative predictors of revision surgery and manipulation under anaesthetic are to be expected, as these indicate that surgery has failed and are indicative of poor outcome. Risk factors such as revision surgery, medical complications and re-operations are uncommon, but should serve as flags to indicate these patients are at risk of chronic pain and may need to be seen by a specialist with knowledge of pain prevention and management. Patients using opioid medication prior to knee replacement are at increased risk of post-operative complications including opioid overdose [37, 38] and have been the subject of increasing attention [39]. Opioids are commonly used in the management of pain both before and after knee replacement despite a lack of recommendation for the use of any types of opioids other than Tramadol, and the evidence of complications [40, 41]. Monitoring analgesic use such as opioids post-operatively can be an indicator of persistent pain following arthroplasty [42] in addition to indicating a poor pain outcome [43]. Use of antidepressants is an important factor to explore, because pain and depression are known to be associated and depression may be a key area for intervention that may help to improve pain coping and management [44]. Antidepressants are commonly used in patients with refractory pain or who have a neuropathic component to their pain [43, 45]. Although pre-operative psychological distress is associated with chronic pain, specific psychological risk factors for chronic pain after knee replacement have been identified with possible management strategies indicated [46].

What this study adds

It is still unclear why patients with little knee pain are undergoing knee replacement and there is a need to understand how decisions about such patients are made. Risk factors of smoking and obesity already feature within NICE guidelines [1], but although weight loss is recommended, the guidelines are clear that ‘patient-specific factors (including smoking, obesity and comorbidities) should not be barriers to referral for joint surgery’. Whilst the majority of patients in these groups do achieve good pain outcomes and should not be denied access to care, they are at an increased absolute risk of poor pain outcome. In the three-month window after surgery, there is an opportunity to provide interventions for patients that reduce the risk of poor pain outcome. Opioid medication prescribing represents an important target for future interventions. Joint replacement is effective but there is a need to focus on careful patient selection and modifiable risk factors to reduce poor outcomes. For those whose outcomes can be optimised it is timely to explore interventions that target such risk factors both before and after surgery.

Imputation model checks with descriptive statistics and univariable model logistic regression results comparing complete case to imputed data.

(DOCX)

Click here for additional data file.

Receiver operating characteristic curve (ROC) curve for discriminatory ability of variables in the final model.

(DOCX)

Click here for additional data file.

14 Oct 2021

PONE-D-21-27576Risk factors associated with poor pain outcomes following primary knee replacement surgery:analysis of data from the Clinical Practice Research Datalink, Hospital Episode Statistics and Patient Reported Outcomes as part of the STAR research programmePLOS ONE

Dear Dr. Judge,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Please submit your revised manuscript by Nov 28 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols.

We look forward to receiving your revised manuscript.

Kind regards,

Armando Almeida

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. We note that the grant information you provided in the ‘Funding Information’ and ‘Financial Disclosure’ sections do not match.

When you resubmit, please ensure that you provide the correct grant numbers and funders name for the awards you received for your study in the ‘Funding Information’ section.

3. Thank you for stating the following in the Competing Interests section:

“I have read the journal's policy and the authors of this manuscript have the following competing interests:

Competing interest statement

All authors have completed the Unified Competing Interest form at www.icmje.org/coi_disclosure.pdf and declare: AJ reports grants from NIHR PGfAR, during the conduct of the study; personal fees from Freshfields Bruckhaus Derringer, personal fees from Anthera Pharmaceuticals Ltd, outside the submitted work; RGH reports grants from NIHR PGfAR, during the conduct of the study. HRM reports grants from Royal College of Surgeons Research Fellowship, outside the submitted work; RPV reports research funding from UK-NIHR, Kyowa Kirin Services, International Osteoporosis Foundation, and lecture fees and/or consulting honoraria from Amgen, UCB, Kyowa Kirin Services, and Mereo Biopharma, all outside of the scope of this study; NA reports personal fees from Pfizer/Lilly, personal fees from Bristows LLP, grants from Merck Grant,  outside the submitted work; VW reports grants from NIHR, during the conduct of the study; TJP reports grants from UK NIHR Programme Grant for Applied Research, during the conduct of the study. All other authors declare no conflicts of interest.”

Please confirm that this does not alter your adherence to all PLOS ONE policies on sharing data and materials, by including the following statement: "This does not alter our adherence to  PLOS ONE policies on sharing data and materials.” (as detailed online in our guide for authors http://journals.plos.org/plosone/s/competing-interests).  If there are restrictions on sharing of data and/or materials, please state these. Please note that we cannot proceed with consideration of your article until this information has been declared.

Please include your updated Competing Interests statement in your cover letter; we will change the online submission form on your behalf.

4. In your Data Availability statement, you have not specified where the minimal data set underlying the results described in your manuscript can be found. PLOS defines a study's minimal data set as the underlying data used to reach the conclusions drawn in the manuscript and any additional data required to replicate the reported study findings in their entirety. All PLOS journals require that the minimal data set be made fully available. For more information about our data policy, please see http://journals.plos.org/plosone/s/data-availability.

Upon re-submitting your revised manuscript, please upload your study’s minimal underlying data set as either Supporting Information files or to a stable, public repository and include the relevant URLs, DOIs, or accession numbers within your revised cover letter. For a list of acceptable repositories, please see http://journals.plos.org/plosone/s/data-availability#loc-recommended-repositories. Any potentially identifying patient information must be fully anonymized.

Important: If there are ethical or legal restrictions to sharing your data publicly, please explain these restrictions in detail. Please see our guidelines for more information on what we consider unacceptable restrictions to publicly sharing data: http://journals.plos.org/plosone/s/data-availability#loc-unacceptable-data-access-restrictions. Note that it is not acceptable for the authors to be the sole named individuals responsible for ensuring data access.

We will update your Data Availability statement to reflect the information you provide in your cover letter.

5. Your ethics statement should only appear in the Methods section of your manuscript. If your ethics statement is written in any section besides the Methods, please move it to the Methods section and delete it from any other section. Please ensure that your ethics statement is included in your manuscript, as the ethics statement entered into the online submission form will not be published alongside your manuscript.

6. Please include captions for your Supporting Information files at the end of your manuscript, and update any in-text citations to match accordingly. Please see our Supporting Information guidelines for more information: http://journals.plos.org/plosone/s/supporting-information.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Partly

Reviewer #2: Yes

Reviewer #3: Yes

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: No

Reviewer #2: Yes

Reviewer #3: I Don't Know

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: No

Reviewer #3: No

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: The authors present a study with a large sample size focusing on risk factors for poor pain outcomes after joint replacement. This is a relevant field of research, but it is my opinion that this manuscript would benefit from a clearer account of the state of the art, to show the readership why this is a novel and relevant study. Additionally, the presentation of results could also be more informative and complete. More specific comments are detailed below.

INTRODUCTION

- Globally, it is my opinion that the introduction section could be more complete, to provide an overview of the state-of-the-art in this field and to show the readership why the present study is warranted. What risk factors for poor outcomes have been analyzed in previous studies? Have the results been consistent across studies? Which risk factors are the most important? Which ones should be further investigated? It is not clear what the present study might add as novelty to the field.

- 1st paragraph, line 1: Should be “have”

- The authors state that “Identifying potentially modifiable risk factors might enable innovation and interventions to reduce the numbers of future patients who experience pain after surgery.” However, few of the analyzed risk factors are modifiable. How does this statement relate to the rest of the work?

METHODS

- It would be more accurate to use the subtitle “Sample” instead of “Population”.

- The variable IMD should be explained since this acronym is not known across all countries.

- In the same line, the Charlson Comorbidity Index should also be briefly explained.

- Why were these specific variables included in the model? There is no information concerning multicollinearity.

Statistical analysis

- Line 1: Patients’

- What is the rationale to categorize the OKS score into 6 categories?

- What is the rationale to exclude underweight BMI?

RESULTS

*Page 6, Pre-operative predictors

- Table 1: Why was the column “No” only filled in for some variables? Using age as an example, it would be possible to report the number of patients with poor outcomes who were “not” in the <60 group. Presumably, this would be the sum of patients who had a poor outcome and were in the other age groups.

- What was the rationale to choose the reference category for each variable? For example, Pre-op OKS score groups are all only compared with the 8-10 score, age groups are all just compared against the <60 group…

- For age, it is stated that “those aged between 60-79 had a reduced risk of poor pain outcome compared with the oldest and youngest age groups”. However, there is no 60-79 group and these age groups (60-69 and 70-79 are not compared with older age groups (only with <60). Can the authors please clarify what is meant?

- I suggest rewriting the sentence “Poor pain outcomes were more likely in current smokers, males, people living in the most deprived areas and those with inflammatory arthritis”. As is, it seems to the reader that the authors are establishing these predictors as the most important ones. Also, I do not believe that these conclusions can be totally inferred from Table 1. I would suggest completing this table with additional statistical information for each predictor.

- I could not find results concerning inflammatory arthritis in the tables or graphs.

- For the remaining predictors, please note that, according to figure 3, effect is are non-significant for Charlson (95% CI=0.99, 1.90), Uni vs. Total arthroplasty (95% CI=0.58, 1.04) and Antidepressants (95% CI=0.98, 1.45) (the CIs include 1).

*Page 7, Absolute risk differences:

- Figure 4: This graph would be more informative if the ARD and CI values were detailed for each predictor, so that the reader can judge the statistical significance of each one.

- What criteria were used to determine what is a “large” ARD?

- When the CI are presented in brackets, the information that the values are the 95% CI should be present, and not only the numbers (e.g., 95% CI: 7.3,16.6)

- Did the authors consider presenting the ROC curve for the model? I believe this would provide a more visual idea of the model.

Page 8, Pain state change:

- “most likely to not improve” – It seems that Table 2 only provides descriptive information. It is not clear how the authors ascertained which patients were most likely to improve or which statistical inference test was used to reach this conclusion.

- “we characterized the pre-operative and post-operative characteristics of those with worsening pain state change.” – Did the authors conduct any statistical test to compare the groups? These conclusions cannot be inferred from descriptive statistics alone. The complete results and p values should be detailed to the reader.

- “There were large differences in” – Was there a formal effect size calculation for between-group differences?

DISCUSSION

- “it may be worthwhile providing pre-surgical interventions to address modifiable risk factors (smoking, obesity and comorbidities)”. According to the results, there is a reduced risk of poor outcomes for yes vs. non smokers, but not for ex vs. non smokers. Therefore, this study does not seem to support pre-surgical interventions to address smoking. Also, the analysis of comorbidities did not show significant results (all CI include 1). Though the identification of risk factors is important, this study mainly focuses on non-modifiable characterizes and thus the clinical implication should be reconsidered.

- Most other studies have looked at composite outcomes combining symptoms of pain, stiffness and function.” – I believe that this may be an overstatement, since many studies in this field analyze pain as an outcome. Please reconsider if this is accurate information.

Reviewer #2: This manuscript reports the results of a study that uses an existing dataset to test for predictors for poor pain outcomes following total knee replacement (TKR). Such studies are common in the field of TKR, but with varying predictors, outcomes and methods. This study is a useful addition to the current body of evidence although, for a study that uses an existing dataset and fairly routine statistical methods, it is surprising to see 16 authors listed. Specific comments follow but these are minor comments and I have no major problems with the paper which is well written, well presented and well conducted.

1. I think the abstract should contain the time point of the outcome

2. The study involves data linkage. This is a process commonly associated with surprisingly high error rates. Some data on the matching rates and processes would be useful.

3. What was done about bilateral procedures – they should either have been excluded from the dataset or included in the model?

4. Population selection: can any information be provided on the likely representativeness of this sample to the population as a whole? I know this is addressed in the discussion but this is based on previous analyses, not this exact dataset.

5. The study reports relative risk ratios (instead of the usual odds ratios generated by logistic regression). Just checking that this is correct – that the authors converted the ORs to RRs?

6. Using the Oxford PS (pain score) as a predictor seems odd when this is the score used to define (calculate) the outcome. I realise that the Oxford PS score and the treatment effect (based on that score) are different but surely those with a worse pre-operative pain score will tend to have larger treatment effects because they have more “room” to improve? This is why I don’t like using treatment effect – I would rather know how much pain they have at 6 months. Pre-operative pain can be added to the model to adjust for its effect. But I am open to arguments to the contrary.

7. The pre-op associations were consistent with other studies and not surprising (except the association with prior knee arthroscopy, which was interesting, and gender was the opposite of what I have seen before). I have a problem with using post-operative data to predict early post-operative outcomes. For example, those taking opioids post-operatively were more likely to have poor pain outcomes. These two variables are kind of measuring the same thing: post-op pain, and it is unlikely that ceasing these drugs will prevent pain at 6 months. Similarly, those who needed further surgery and had complications were more likely to have pain. These findings are expected and don’t point to any clinically useful knowledge, apart from avoiding complications. I am not asking the authors to remove them, but some comment in the discussion about the limited usefulness and obviousness of these findings should be mentioned.

Ian Harris

Professor of Orthopaedic Surgery, UNSW Sydney

Reviewer #3: The manuscript reports the findings from an observational study aiming to identify risk factors for poor pain outcomes after TKR. Pain outcomes were defined using the pain section of the OKS.

ABSTRACT

In the manuscript, complications seem important, but are not mentioned here?

INTRODUCTION

1. Suggest modify sentence – ‘but around 1 in 5 will continue to experience pain…’ The 20% isn’t upheld in many studies. Suggest change to “the % experience ongoing pain is variable (add refs) with up to 20% experiencing…”. The current data also supports the finding that 20% is not often upheld.

(Suggest the authors could do a systematic review on whether the % with persistent pain has improved across time? (as another study) as this may explain why 20% seems an outdated value now)

METHODS

1. The introduction talks about TKA for people with OA. Clarify if only people with OA are in the dataset used. If not, change the Intro to be more inclusive of other indications for surgery.

2. Justify inclusion of unicompartmental surgery. Its inclusion implies the same predictors will apply

3. Please elaborate on justification for included complications. Were these defined by stakeholders? Were they chosen on severity? This is important as you include UTI (minor) to most severe and transparency is required here.

4. Clarify total number of GP visits? From surgery to 3 months post-surgery

5. Mention model fit statistics and performance test to be used here (mentioned in Results)

6. Clarify medication use includes pre and post-op. It gets confusing when talking about med use in Results.

RESULTS

1. Missing data for alcohol consumption of 17% is high. Should do sensitivity analysis without that variable otherwise justify why that is not necessary.

2. Clarify you checked for correlations between medication use pre and post-op. Same with opioid use and pain pre and post-op? There may be collinearity there.

Table 1

Clarify that ‘complication’ is different to say surgery for MUA? I would argue MUA is a subset of complication (as this applies to some other complications too). By keeping them separate, this assumes different complications have different ‘weights’ so to speak. Can you justify/explain this approach. Seems like you have distinguished ‘medical’ from surgical and surgical is broken down further?

Fig 1 - Please justify exclusion of underweight BMI

DISCUSSION

1. Discussing use of opioids post-operatively as a risk factor along side pre-op factors is confusing. It makes sense that opioid use is assoc with persistent pain if the pain is driving use. On the other hand, pre-op BMI as a predictor is completely different. It may be a predictor as opposed to opioid use which may not ‘predict’, but rather be reactionary.

Can the authors try to tease this out or dela with this better.

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: Yes: Ian A Harris

Reviewer #3: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

17 Nov 2021

Dear Dr Almeida,

Thank you for reviewing our manuscript at PLOS ONE as per your email date 15 Oct 2021. We have considered these comments carefully, and our responses and amendments to the manuscript are reported below.

1. Reviewer #1: The authors present a study with a large sample size focusing on risk factors for poor pain outcomes after joint replacement. This is a relevant field of research, but it is my opinion that this manuscript would benefit from a clearer account of the state of the art, to show the readership why this is a novel and relevant study. Additionally, the presentation of results could also be more informative and complete. More specific comments are detailed below.

Thank you for your comments. We have revised the manuscript to give a clearer account of current knowledge, describe our study and improved the presentation of results.

[see Introduction and Results, and below]

2. INTRODUCTION - Globally, it is my opinion that the introduction section could be more complete, to provide an overview of the state-of-the-art in this field and to show the readership why the present study is warranted. What risk factors for poor outcomes have been analyzed in previous studies? Have the results been consistent across studies? Which risk factors are the most important? Which ones should be further investigated? It is not clear what the present study might add as novelty to the field.

Most previous studies have focused on pre-operative risk factors when assessing predictors of patients reported outcomes. Many studies have looked at composite outcomes that combine information on pain, stiffness and function, whereas patients tell us that the most important outcome to them is pain. Our previous research found that the pre-operative risk factors were different for pain and functional domains. In particular, the discriminatory ability of the models to predict function were much better, than for pain outcome, where the discriminatory ability of pre-operative risks factors was poor (https://pubmed.ncbi.nlm.nih.gov/22532699/). Within the literature less is known about whether post-operative risk factors can help improve prediction.

The specific aim of our study was to use a wide and comprehensive range of information on predictors, that covered both pre and post-operative information, to see if this could help improve our prediction and better identify patients at risk of post-surgical chronic pain.

We have now expanded the introduction to provide a more complete overview of the rationale for this study.

[Introduction, paragraph 2]

3. - 1st paragraph, line 1: Should be “have”

Amended [Introduction, paragraph 1]

4. - The authors state that “Identifying potentially modifiable risk factors might enable innovation and interventions to reduce the numbers of future patients who experience pain after surgery.” However, few of the analyzed risk factors are modifiable. How does this statement relate to the rest of the work?

We agree with the comment and have removed this sentence.

[Introduction, paragraph 2]

5. METHODS

- It would be more accurate to use the subtitle “Sample” instead of “Population”.

Amended [Methods, paragraph 3]

6. - The variable IMD should be explained since this acronym is not known across all countries.

We agree and have added the following sentences:

To measure socioeconomic deprivation, we used the index of multiple deprivation (IMD) score. This is a relative measure of deprivation for small areas – termed lower super output areas (LSOAs) – which are defined as geographical areas of a similar population size, with an average of 1,500 residents [19]. The IMD comprises seven measures of deprivation: income deprivation; employment deprivation; education, skills and training deprivation; health deprivation and disability; crime; barriers to housing and services; and living environment deprivation. We used the IMD rank for a patient’s LSOA and categorised patients into quintiles based upon the national ranking of local areas, with quintile 1 being the least deprived group and quintile 5 being the most deprived group (i.e. reordered to aid reporting).

[Methods, Pre-operative predictors]

7. - In the same line, the Charlson Comorbidity Index should also be briefly explained.

We agree and have added the following sentences:

As a measure of comorbidity we used the Royal College of Surgeons’ (RCS) Charlson Score, which is calculated based on the presence of several chronic conditions identified using ICD-10 codes at the time of knee replacement surgery admission and all admissions in the preceding 5 years [20].

[Methods, Pre-operative predictors]

8. - Why were these specific variables included in the model? There is no information concerning multicollinearity.

In discussion with clinicians we sought to identify variables within this routinely collected dataset which may provide a wide range of potential risk factors/predictors for the pain outcome. We have assessed for evidence of collinearity using variance inflation factors and there was no evidence of multicollinearity. We have added statements regarding both points.

Predictor variables: In discussion with clinicians we sought to identify variables within this routinely collected dataset which may provide a wide range of potential predictors for the pain outcome.

We have assessed for evidence of collinearity using variance inflation factors and there was no evidence of multicollinearity.

[Methods, Predictor variables and paragraph 11]

9. Statistical analysis – Line 1: Patients’

Amended [Statistical analysis, paragraph 1]

10. - What is the rationale to categorize the OKS score into 6 categories?

One of the assumptions of a regression model is linearity of continuous variables with the outcome – we used fractional polynomial regression models to assess this and there was strong evidence of non-linearity of pre-op knee scores, so to satisfy the assumption we categorised the OKS score.

11. - What is the rationale to exclude underweight BMI?

We have added a sentence to clarify this point:

We excluded nine underweight BMI patients, as there were too few patients in this category, leading to a small cell problem in the multivariable regression model, and it would be inappropriate to combine underweight and normal BMI categories together.

[Statistical analysis, paragraph 2]

12. RESULTS Page 6, Pre-operative predictors – Table 1: Why was the column “No” only filled in for some variables? Using age as an example, it would be possible to report the number of patients with poor outcomes who were “not” in the <60 group. Presumably, this would be the sum of patients who had a poor outcome and were in the other age groups.

To clarify Table 1 column headings No and Yes refer to whether the risk factor is present or not, rather than whether pain is present or not. For example, for the hypertension row the No column refers to those with no hypertension who have pain. The Yes column those with hypertension and pain. Therefore, for those aged <60 years with pain (Yes column=15.7%), the No column would have to contain all those that are >=60 years with pain. We could remove the No column and add two rows per binary outcome (example below). However, for now we have left the table unchanged. We will revise the table should you feel strongly that it should be changed.

13. - What was the rationale to choose the reference category for each variable? For example, Pre-op OKS score groups are all only compared with the 8-10 score, age groups are all just compared against the <60 group…

Our rational for choosing the reference category is the usual practice of selecting the group with the largest sample size, as our dataset is fairly large, we selected the lowest category for each variable, we are happy to change reference categories if you feel strongly regarding this point.

14. - For age, it is stated that “those aged between 60-79 had a reduced risk of poor pain outcome compared with the oldest and youngest age groups”. However, there is no 60-79 group and these age groups (60-69 and 70-79 are not compared with older age groups (only with <60). Can the authors please clarify what is meant? The effect size is bigger in the middle age groups

We have amended this sentence to clarify its meaning:

A ‘U-shaped’ effect was observed for patient age at surgery, where those aged between 60-69 and 70-79 had a reduced risk of poor pain outcome compared with the youngest age group. The oldest age group was similar to the youngest.

[Results, paragraph 3]

15. - I suggest rewriting the sentence “Poor pain outcomes were more likely in current smokers, males, people living in the most deprived areas and those with inflammatory arthritis”. As is, it seems to the reader that the authors are establishing these predictors as the most important ones. Also, I do not believe that these conclusions can be totally inferred from Table 1. I would suggest completing this table with additional statistical information for each predictor.

At this stage of the results we are examining the descriptive statistics, looking at associations in the data and we are highlighting those that appear the most clinically relevant. There is no formal hypothesis testing being done, this is done with the regression modelling. Please see earlier response about missing data in Table 1. We have rewritten the sentence: The descriptive associations suggest poor pain outcomes were more likely in current smokers, males, people living in the most deprived areas and those with inflammatory arthritis (Table 1).

[Results, paragraph 3]

16. - I could not find results concerning inflammatory arthritis in the tables or graphs.

Inflammatory arthritis appears in Table 1 below the subheading ‘Comorbidities’. No individual comorbidities appear in the forest plot as they are rare, some of the categories are too small so we present findings for combined comorbidities in the Charleston score. We have added a sentence to the methods to clarify this:

Composite variables were used if individual characteristics were rare.

[Statistical analysis, paragraph 2]

17. - For the remaining predictors, please note that, according to figure 3, effect is are non-significant for Charlson (95% CI=0.99, 1.90), Uni vs. Total arthroplasty (95% CI=0.58, 1.04) and Antidepressants (95% CI=0.98, 1.45) (the Cis include 1).

Please see paper by Sterne et al BMJ (https://www.bmj.com/content/322/7280/226.1.full). We are examining the strengths of association and not arbitrary measures of statistical significance with cut offs of for example p<0.05 or the related concept of whether the confidence interval includes the null value. To clarify we have added this sentence:

For the following variables there is weak evidence of an association.

[Results, paragraph 3]

18. *Page 7, Absolute risk differences:

- Figure 4: This graph would be more informative if the ARD and CI values were detailed for each predictor, so that the reader can judge the statistical significance of each one.

Thank you we agree, and have added the absolute risk difference and confidence intervals to Figure 4.

19. - What criteria were used to determine what is a “large” ARD?

We did not use any specific criteria, rather this was in respect of ‘strength of association’ as per the Bradford Hill criteria. We have now amended the wording: The other pre-operative variables with some absolute adjusted risk differences were…

[Results, Absolute risk differences]

20. - When the CI are presented in brackets, the information that the values are the 95% CI should be present, and not only the numbers (e.g., 95% CI: 7.3,16.6)

Amended [Abstract and Results, paragraph 6]

21. - Did the authors consider presenting the ROC curve for the model? I believe this would provide a more visual idea of the model.

Thank you we agree and have now added the ROC curve to the Supplementary data file and Results, paragraph 7.

22. Page 8, Pain state change:

- “we characterized the pre-operative and post-operative characteristics of those with worsening pain state change.” – Did the authors conduct any statistical test to compare the groups? These conclusions cannot be inferred from descriptive statistics alone. The complete results and p values should be detailed to the reader.

This is descriptive not hypothesis testing so we have amended the language and made it clear in the methods and results.

To describe their change in pain state before and 6 months after surgery, patients are categorised into pain groups, and descriptive statistics (without statistical testing) used to describe the number (percentage) of patients that move between different pain states before and after surgery.

… we describe the pre-operative and post-operative characteristics of those with worsening pain state change.

[Statistical analysis, paragraph 1 and Results, Pain state change]

23. - “There were large differences in” – Was there a formal effect size calculation for between-group differences?

We agree and have amended wording:

There were some differences in length of stay, where those who moved to a worse pain state had much shorter length of stay

[Results, Pain state change]

24. DISCUSSION – “it may be worthwhile providing pre-surgical interventions to address modifiable risk factors (smoking, obesity and comorbidities)”. According to the results, there is a reduced risk of poor outcomes for yes vs. non smokers, but not for ex vs. non smokers. Therefore, this study does not seem to support pre-surgical interventions to address smoking. Also, the analysis of comorbidities did not show significant results (all CI include 1). Though the identification of risk factors is important, this study mainly focuses on non-modifiable characterizes and thus the clinical implication should be reconsidered.

We agree and have removed the statement.

[Discussion, paragraph 5-What this study adds]

25. - Most other studies have looked at composite outcomes combining symptoms of pain, stiffness and function.” – I believe that this may be an overstatement, since many studies in this field analyze pain as an outcome. Please reconsider if this is accurate information.

Scores like the Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) and Oxford Knee Score are commonly used composite outcomes. We have amended wording to clarify our meaning.

Some other studies have looked at composite outcomes

[Discussion, paragraph 3-What is already known]

26. Reviewer #2: This manuscript reports the results of a study that uses an existing dataset to test for predictors for poor pain outcomes following total knee replacement (TKR). Such studies are common in the field of TKR, but with varying predictors, outcomes and methods. This study is a useful addition to the current body of evidence although, for a study that uses an existing dataset and fairly routine statistical methods, it is surprising to see 16 authors listed. Specific comments follow but these are minor comments and I have no major problems with the paper which is well written, well presented and well conducted.

27. 1. I think the abstract should contain the time point of the outcome

Amended the objective to: Identify risk factors for poor pain outcomes six months after primary knee replacement surgery.

[Abstract]

28. 2. The study involves data linkage. This is a process commonly associated with surprisingly high error rates. Some data on the matching rates and processes would be useful.

The data linkage is done in-house by CPRD and not by the researchers themselves. Providers of data such as CPRD and NHS Digital provide the service for data linkage as a ‘trusted third party’ in a secure environment. This ensures the researchers themselves only have access to the anonymised linked data, and for information governance reasons, there is no need for sharing of personal data. Information on CPRD linkage is available here https://cprd.com/linked-data.

Hence researchers who use data provided by CPRD do not have information on the data linkage process itself.

We now describe in the methods of the paper that: Linkage of CPRD-HES-ONS-PROMS data is done by NHS Digital as a ‘trusted third party’.

[Methods, paragraph 2]

29. 3. What was done about bilateral procedures – they should either have been excluded from the dataset or included in the model?

Within this routine dataset, we do not have information on whether or not a patient received a unilateral or bilateral knee replacement. This is now described as a limitation in the discussion section.

However, within this routine dataset, we do not have information on whether a patient received a unilateral or bilateral knee replacement and hence we are unable to exclude bilateral procedures from the dataset.

[Discussion, Strengths and limitations]

30. 4. Population selection: can any information be provided on the likely representativeness of this sample to the population as a whole? I know this is addressed in the discussion but this is based on previous analyses, not this exact dataset.

Data from the national joint registry provides information on patients receiving hip and knee replacement where there is mandatory data collection nationally. To address the reviewers question, we have compared the mean age, and sex of patients in our CPRD sample, to that of the NJR https://reports.njrcentre.org.uk/Portals/0/PDFdownloads/NJR%2018th%20Annual%20Report%202021.pdf. For knee replacement in NJR mean age 68.9 (SD 9.6), 56.6% female. In our CPRD sample mean age of 69 (SD 9) of whom 56.1% were female. Hence very comparable in respect of age and gender. This has been added in the discussion section.

This sample has also been compared with the mandatory National Joint Registry (NJR) in respect to knee replacement patient profile, for NJR mean age 68.9 (SD 9.6), 56.6% female. In our CPRD sample, the mean age was 69 (SD 9), of whom 56.1% were female.

[Discussion, Strengths and limitations]

31. 5. The study reports relative risk ratios (instead of the usual odds ratios generated by logistic regression). Just checking that this is correct – that the authors converted the Ors to RRs?

We confirm that in this study we are indeed using relative risk ratios. Logistic regression is appropriate to use for relatively infrequent outcomes, whereas for more common outcomes, odds ratios are not a good approximation for relative risk. Hence in this paper, we have presented the results as relative risk ratios by fitting a generalized linear model with a binomial error structure and a log link function (log-logistic model) in order to estimate the relative risk.

32. 6. Using the Oxford PS (pain score) as a predictor seems odd when this is the score used to define (calculate) the outcome. I realise that the Oxford PS score and the treatment effect (based on that score) are different but surely those with a worse pre-operative pain score will tend to have larger treatment effects because they have more “room” to improve? This is why I don’t like using treatment effect – I would rather know how much pain they have at 6 months. Pre-operative pain can be added to the model to adjust for its effect. But I am open to arguments to the contrary.

We understand the point that the reviewer is making, and there is some debate in the literature as to how to define patient reported outcomes in observational studies with repeated measures repeated before and after surgery. This is discussed for example, in the article by Losina and Katz, as to whether interest is in the ‘journey’ (that is, absolute or relative change in pain score before and after surgery), or the ‘destination’ (that is, the attained post-operative level of pain). https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3448885/

The limitation with a continuous OKS pain score outcome adjusting for baseline OKS as a covariate (such as in analysis of covariance ANCOVA models) is that, for knee replacement, the majority of patients achieve substantial improvements in pain, and we end up identifying predictors of patients with excellent improvement in pain, compared with patients receiving really good improvements of pain https://online.boneandjoint.org.uk/doi/pdf/10.1302/0301-620X.94B3.27425. Whereas our actual interest is in trying to identify the minority of patients at the tail of the distribution who do not achieve good improvement in symptoms, and hence to define a binary outcome according to whether or not patients had a good pain outcome after surgery. Our preference is to use the treatment effect to define outcome, and we have previously reported on this approach and validated it (Huber J, Husler J, Dieppe P, et al. A new responder criterion (relative effect per patient (REPP) > 0.2) externally validated in a large total hip replacement multicenter cohort (EUROHIP). Osteoarthritis and Cartilage 2016;24(3):480-3.)

The rationale for our descriptive analyses, looking at pain state change, was to better understand the comment made by the reviewer, that patients with larger treatment effects have more room to improve. We noted that patients with the mildest pre-operative pain symptoms were most likely to not improve and their pain actually got worse. Specifically, they moved to a worse pain state following surgery, where 20% of patients had a poor pain outcome compared with around 10% in the other pre-operative pain states. We characterised why this might be, where these patients with mild pre-operative symptoms who experienced worse outcomes had more complications and re-operations.

33. 7. The pre-op associations were consistent with other studies and not surprising (except the association with prior knee arthroscopy, which was interesting, and gender was the opposite of what I have seen before). I have a problem with using post-operative data to predict early post-operative outcomes. For example, those taking opioids post-operatively were more likely to have poor pain outcomes. These two variables are kind of measuring the same thing: post-op pain, and it is unlikely that ceasing these drugs will prevent pain at 6 months. Similarly, those who needed further surgery and had complications were more likely to have pain. These findings are expected and don’t point to any clinically useful knowledge, apart from avoiding complications. I am not asking the authors to remove them, but some comment in the discussion about the limited usefulness and obviousness of these findings should be mentioned.

This study has been conducted as part of a wider NIHR programme grant on chronic knee pain, where the a priori aim was to see if identification and inclusion of post-operative risk factors, could help us to identify patients who are most at risk of chronic post operative pain, rather than using pre-operative risk factors alone. The funding application for the NIHR programme grant went through a rigorous process of peer review, and this is the research question we have been funded to address.

The reviewers’ points relating to opioid use, early complications, and further surgery, are well made, and we have included this within the discussion section. However, knowledge of such post-operative factors would help us to identify patients earlier, even if by a few months, allowing us to intervene earlier and ensure such patients are receiving appropriate pain management services.

34. Ian Harris Professor of Orthopaedic Surgery, UNSW Sydney

Reviewer #3: The manuscript reports the findings from an observational study aiming to identify risk factors for poor pain outcomes after TKR. Pain outcomes were defined using the pain section of the OKS.

35. ABSTRACT

In the manuscript, complications seem important, but are not mentioned here?

These have now been included in the abstract.

Those patients with worsening pain state change had more complications by 3-months (11.8% among those in a worse pain state vs. 2.7% with the same pain state).

36. INTRODUCTION

1. Suggest modify sentence – ‘but around 1 in 5 will continue to experience pain…’ The 20% isn’t upheld in many studies. Suggest change to “the % experience ongoing pain is variable (add refs) with up to 20% experiencing…”. The current data also supports the finding that 20% is not often upheld.

The figure of 20% was determined through a systematic review of the literature by Beswick et al (https://pubmed.ncbi.nlm.nih.gov/22357571/). We acknowledge the reviewer’s point that this was published some time ago now, and that this estimate will vary between studies and change over time. Hence, we have modified this statement as suggested.

Many patients can expect to achieve reductions in knee pain and improvements in functional outcomes [3]. The percentage who experience ongoing pain is variable [4], with up to 20% experiencing knee pain that impacts their quality of life [5].

[Introduction, paragraph 1]

37. (Suggest the authors could do a systematic review on whether the % with persistent pain has improved across time? (as another study) as this may explain why 20% seems an outdated value now)

We agree with the reviewer that this will be a useful area for a future study, and have mentioned this in the discussion section, but is beyond the scope of this particular paper.

38. METHODS

1. The introduction talks about TKA for people with OA. Clarify if only people with OA are in the dataset used. If not, change the Intro to be more inclusive of other indications for surgery.

The reviewer is correct that this paper contains patients receiving a primary elective knee replacement for all indications, and not just osteoarthritis. We have changed the introduction to be more inclusive of other indications.

Over 100,000 knee replacement operations are carried out each year in the UK for osteoarthritis and other surgical indications [2]

[Introduction, paragraph 1]

39. 2. Justify inclusion of unicompartmental surgery. Its inclusion implies the same predictors will apply We included all patients receiving knee replacement surgery, whether total or uni-compartmental.

We included total versus uni-compartmental as a predictor in the model, to see if this was associated with worse pain outcomes, with weak evidence of unicompartmental patients having lower risk of worse pain outcomes.

A limitation of this, as the reviewer highlights, is that we are making an assumption - namely, that risks factors of poor pain outcomes, are the same for patients receiving uni-compartmental and total knee replacement. Investigating this would require tests for interactions, with all other risk factors in the model, but such multiple testing could lead to type 1 errors being made, and are in any case very low powered.

We have therefore made it clear in the methods that the population comprises patients receiving primary total and uni-compartmental knee replacement [Methods, Sample paragraph 3]. In the discussion, we describe this limitation, and the assumption being made for other risk factors identified.

Another limitation is that we are making an assumption that risk factors of poor pain outcomes, are the same for patients receiving uni-compartmental and total knee replacement. Testing for this would require test for interaction, with all other risk factors in the model, but such multiple testing could lead to type 1 errors and are in any case very low powered.

[Discussion, Strengths and limitations]

40. 3. Please elaborate on justification for included complications. Were these defined by stakeholders? Were they chosen on severity? This is important as you include UTI (minor) to most severe and transparency is required here.

Complications were chosen a-priori in discussion with a number of orthopaedic surgeons, as to what were considered to be clinically relevant complications of knee replacement surgery. The chosen complications have been described and published previously https://www.oarsijournal.com/article/S1063-4584(14)01051-6/fulltext

We have added this reference to the Methods, Post-operative predictors.

41. 4. Clarify total number of GP visits? From surgery to 3 months post-surgery

Amended

and calculated the total number of general practice visits between surgery and 3 months post-surgery

[Methods, paragraph 11]

42. 5. Mention model fit statistics and performance test to be used here (mentioned in Results)

Added to Statistical analysis:

The C-statistic was used to describe the discriminatory ability of variables in the final model.

43. 6. Clarify medication use includes pre and post-op. It gets confusing when talking about med use in Results.

Amended

We identified medications prescribed (including opioids, NSAIDS, and antibiotics) pre- and post-surgery

[Methods, paragraph 11]

44. RESULTS

1. Missing data for alcohol consumption of 17% is high. Should do sensitivity analysis without that variable otherwise justify why that is not necessary.

We have carried out a sensitivity analysis, re-fitting the final model excluding alcohol consumption, and the findings remain unchanged.

45. 2. Clarify you checked for correlations between medication use pre and post-op. Same with opioid use and pain pre and post-op? There may be collinearity there.

We have checked for evidence of multicollinearity using variance inflation factors as part of our model regression diagnostics, and found no evidence of this being an issue. See above point 8.

46. Table 1

Clarify that ‘complication’ is different to say surgery for MUA? I would argue MUA is a subset of complication (as this applies to some other complications too). By keeping them separate, this assumes different complications have different ‘weights’ so to speak. Can you justify/explain this approach. Seems like you have distinguished ‘medical’ from surgical and surgical is broken down further?

In advance, we chose to separate out manipulation under anaesthetic, as this is treated as an operation using OPCS4 codes. Medical and surgical complications were considered as separate predictors. There was no specific reason for the approach we chose, and the reviewer is correct that others may have defined this differently. We have described this as a limitation in the discussion section.

Medical and surgical complications were considered as separate predictors a priori and others may have defined complications differently from this study.

[Discussion, Strengths and limitations]

47. Fig 1 - Please justify exclusion of underweight BMI

See above

48. DISCUSSION

1. Discussing use of opioids post-operatively as a risk factor along side pre-op factors is confusing. It makes sense that opioid use is assoc with persistent pain if the pain is driving use. On the other hand, pre-op BMI as a predictor is completely different. It may be a predictor as opposed to opioid use which may not ‘predict’, but rather be reactionary. Can the authors try to tease this out or dela with this better.

We have tried to clarify our meaning regarding opioid use:

We have identified a number of risk factors that are associated with an increased risk of poor pain outcome. The strongest pre-operative risk factors were: having only mild knee pain symptoms at the time of surgery, being a current smoker, obesity, and living in the most deprived areas. Opioid and antidepressant medication use were also associated with worse pain outcomes. The strongest post-operative factors were revision surgery and manipulation under anaesthetic within three months after the operation. We identified a range of other important risk factors with more moderate effects in terms of absolute risk differences in pain outcome, including a history of previous knee arthroscopy, and use of opioids within the three months after surgery, in addition to a number of other risk factors. Those with the least pre-operative pain were more likely to move to a worse post-operative pain state and were most likely to take pain relieving medicines both pre- and post-surgery, including opioids.

[Discussion, Main findings]

We hope our responses are satisfactory, but please do not hesitate to contact us if more information is required.

Yours sincerely

Professor Andy Judge

Submitted filename: STAR PLOS ONE response to reviewers 17Nov21.docx

Click here for additional data file.

26 Nov 2021

PONE-D-21-27576R1Risk factors associated with poor pain outcomes following primary knee replacement surgery:analysis of data from the Clinical Practice Research Datalink, Hospital Episode Statistics and Patient Reported Outcomes as part of the STAR research programmePLOS ONE

Dear Dr. Judge,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Please submit your revised manuscript by Jan 10 2022 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols.

We look forward to receiving your revised manuscript.

Kind regards,

Armando Almeida

Academic Editor

PLOS ONE

Journal Requirements:

Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #1: (No Response)

Reviewer #2: All comments have been addressed

Reviewer #3: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: Yes

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: I thank the authors for their answers to my previous comments. I have some additional suggestions that are detailed below.

INTRODUCTION

Page 3, first paragraph: “Many patients can expect to achieve reductions in knee pain and improvements in functional outcomes [3]. The percentage who experience ongoing pain is variable [4], with up to 20% experiencing knee pain that impacts their quality of life [5]. Patients who experience this kind of pain after surgery have not received the expected benefit and for some their pain is worse than it was before the operation [6, 7].” – In this paragraph, the time point at which these patients still experience pain is not clear. I can assume that it refers to chronic pain, but it should it should be stated.

Page 3, last paragraph: “There is limited research focusing solely on pain status”. In fact, there is a large body of research focusing on acute and chronic pain after surgery, including systematic reviews. I do not think that this sentence is an accurate depiction of the state of the art.

Page 3, last paragraph: “This is important given that up to 20% of patients will have long-term pain after surgery [5].” – This information is already stated in the previous paragraph.

Page 4: Please consider if the objective would be improved by stating that the aim was to “identify pre and postoperative risk factors”, since the authors present this strategy as a novelty.

RESULTS

Table 1: I appreciate the clarifications made by the authors concerning Table 1. However, it is my opinion that the information in the table is not very intuitive to understand. One suggestion would be to change the heading of the third column to “Proportion of patients with poor pain response with and without each risk factor”. And then change the “Yes” and “No” to “With” and “Without”.

Also, please consider if using the term “risk factor” is adequate in this context. Only the statistical analyses tell us if each characteristic is a risk factor or not (and not the descriptive data).

Since this table only presents descriptive information, do the authors believe that it is accurate to make claims about risk factors based on its information? Probably the most accurate way of stating the results would be, for example “The highest proportion of patients with poor pain outcomes were in the group of current smokers, males, people living in the most deprived areas and those with inflammatory arthritis.”

The rationale to select reference categories should be clear in the statistical analyses section, even if it is usual practice.

The information provided by the authors in the response letter concerning the analysis of strength of association should be stated in the statistical analysis section (We are examining the strengths of association and not arbitrary measures of statistical significance with cut offs of for example p<0.05 or the related concept of whether the confidence interval includes the null value.)

Reviewer #2: Comments addressed satisfactorily

Reviewer #3: The authors have done well to address all the reviewer comments. Some reanalysis has been undertaken and substantial amendments to the manuscript have been made

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: Yes: Ian Harris, Professor of Orthopaedic Surgery, UNSW Sydney

Reviewer #3: Yes: Justine M Naylor

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

9 Dec 2021

Dear Dr Almeida,

Thank you for reviewing our manuscript at PLOS ONE as per your email dated 15 Oct 2021. We have considered these comments carefully, and our responses and amendments to the manuscript are reported below.

Reviewer #1: I thank the authors for their answers to my previous comments. I have some additional suggestions that are detailed below.

Thank you, we are very grateful for your comments.

1. INTRODUCTION

Page 3, first paragraph: “Many patients can expect to achieve reductions in knee pain and improvements in functional outcomes [3]. The percentage who experience ongoing pain is variable [4], with up to 20% experiencing knee pain that impacts their quality of life [5]. Patients who experience this kind of pain after surgery have not received the expected benefit and for some their pain is worse than it was before the operation [6, 7].” – In this paragraph, the time point at which these patients still experience pain is not clear. I can assume that it refers to chronic pain, but it should it should be stated.

Thank you for your comments we agree and have amended these sentences.

Many patients can expect to achieve reductions in knee pain and improvements in functional outcomes following surgery [3]. The percentage who experience ongoing chronic knee pain post-surgery is variable [4], with up to 20% experiencing knee pain that impacts their quality of life after 3 months post-op [5]. Patients who experience this kind of pain after surgery have not received the expected benefit and for some their pain is worse than it was before the operation [6, 7].

[Introduction, paragraph 2]

2. Page 3, last paragraph: “There is limited research focusing solely on pain status”. In fact, there is a large body of research focusing on acute and chronic pain after surgery, including systematic reviews. I do not think that this sentence is an accurate depiction of the state of the art.

We agree and have amended the sentence, to clarify our meaning.

Although previous research has explored predictors of outcomes of knee replacement [8], most studies have focused on total scores encompassing several domains (e.g. pain, stiffness and function) and fewer studies have focussed solely on pain status [7].

[Introduction, paragraph 2]

3. Page 3, last paragraph: “This is important given that up to 20% of patients will have long-term pain after surgery [5].” – This information is already stated in the previous paragraph.

We agree and have removed the sentence.

[Introduction, paragraph 2]

4. Page 4: Please consider if the objective would be improved by stating that the aim was to “identify pre and postoperative risk factors”, since the authors present this strategy as a novelty.

We agree and have added this statement.

The aim of this study is to identify pre- and post-operative risk factors for whether or not a patient has a poor pain outcome after knee replacement surgery…

[Introduction, paragraph 3]

5. RESULTS

Table 1: I appreciate the clarifications made by the authors concerning Table 1. However, it is my opinion that the information in the table is not very intuitive to understand. One suggestion would be to change the heading of the third column to “Proportion of patients with poor pain response with and without each risk factor”. And then change the “Yes” and “No” to “With” and “Without”.

We agree and have amended the table headings.

Proportion of patients with poor pain response with and without each risk factor

Without With

[Table 1]

6. Also, please consider if using the term “risk factor” is adequate in this context. Only the statistical analyses tell us if each characteristic is a risk factor or not (and not the descriptive data).

We agree and have clarified our wording for the descriptive data.

The aim of this study is to identify pre- and post-operative risk factors for whether or not a patient has a poor pain outcome after knee replacement surgery, by analysing a wide range of potential factors from the UK Clinical Practice Research Datalink (CPRD) primary care GOLD database linked to English Hospital Episode Statistics (HES) hospital admissions and to Patient Reported Outcomes Measures (PROMs) data.

[Introduction, paragraph 3]

Table 1. Descriptive statistics describing the total number of patients with each potential risk factor, and the proportion of patients with a poor pain outcome, according to whether or not they have the factor

[Table 1]

7. Since this table only presents descriptive information, do the authors believe that it is accurate to make claims about risk factors based on its information? Probably the most accurate way of stating the results would be, for example “The highest proportion of patients with poor pain outcomes were in the group of current smokers, males, people living in the most deprived areas and those with inflammatory arthritis.”

We agree and have amended this sentence.

The highest proportion of patients with poor pain outcomes were in the group of current smokers, males, people living in the most deprived areas and those with inflammatory arthritis (Table 1).

[Results, paragraph 3]

8. The rationale to select reference categories should be clear in the statistical analyses section, even if it is usual practice.

We agree and have added the following sentence.

As our dataset is large, we selected the lowest category for each variable as the reference category.

[Statistical analysis, paragraph 2]

9. The information provided by the authors in the response letter concerning the analysis of strength of association should be stated in the statistical analysis section (We are examining the strengths of association and not arbitrary measures of statistical significance with cut offs of for example p<0.05 or the related concept of whether the confidence interval includes the null value.)

We agree and have added this sentence.

We examine the strength of associations and not arbitrary measures of statistical significance with cut offs of for example p<0.05 or the related concept of whether the confidence interval includes the null value.

[Statistical analysis, paragraph 2]

Reviewer #2: Comments addressed satisfactorily

Thank you.

Reviewer #3: The authors have done well to address all the reviewer comments. Some reanalysis has been undertaken and substantial amendments to the manuscript have been made

Thank you.

We hope our responses are satisfactory, but please do not hesitate to contact us if more information is required.

Yours sincerely

Professor Andy Judge

Submitted filename: STAR PLOS ONE response to reviewers 2nd review 6Dec21 AJ.docx

Click here for additional data file.

13 Dec 2021

Risk factors associated with poor pain outcomes following primary knee replacement surgery:analysis of data from the Clinical Practice Research Datalink, Hospital Episode Statistics and Patient Reported Outcomes as part of the STAR research programme

PONE-D-21-27576R2

Dear Dr. Judge,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Armando Almeida

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

Reviewers' comments:

20 Dec 2021

PONE-D-21-27576R2

Risk factors associated with poor pain outcomes following primary knee replacement surgery:analysis of data from the Clinical Practice Research Datalink, Hospital Episode Statistics and Patient Reported Outcomes as part of the STAR research programme

Dear Dr. Judge:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Prof. Armando Almeida

Academic Editor

PLOS ONE