Exploring red cell distribution width as a potential risk factor in emergency bowel surgery-A retrospective cohort study.

Increased preoperative red cell distribution width (RDW) is associated with higher mortality following non-cardiac surgery in patients older than 65 years. Little is known if this association holds for all adult emergency laparotomy patients and whether it affects 30-day or long-term mortality. Thus, we examined the relationship between increased RDW and postoperative mortality. Furthermore, we investigated the prognostic worth of anisocytosis and explored a possible association between increased RDW and frailty in this cohort. We conducted a retrospective, single centre National Emergency Laparotomy Audit (NELA) database study at St Mary's Hospital Imperial NHS Trust between January 2014 and April 2018. A total of 356 patients were included. Survival models were developed using Cox regression analysis, whereas RDW and frailty were analysed using multivariable logistic regression. Underlying model assumptions were checked, including discrimination and calibration. We internally validated our models using bootstrap resampling. There were 33 (9.3%) deaths within 30-days and 72 (20.2%) overall. Median RDW values for 30-day mortality were 13.8% (IQR 13.1%-15%) in survivors and 14.9% (IQR 13.7%-16.1%) in non-survivors, p = 0.007. Similarly, median RDW values were lower in overall survivors (13.7% (IQR 13%-14.7%) versus 14.9% (IQR 13.9%-15.9%) (p<0.001)). Mortality increased across quartiles of RDW, as did the proportion of frail patients. Anisocytosis was not associated with 30-day mortality but demonstrated a link with overall death rates. Increasing RDW was associated with a higher probability of frailty for 30-day (Odds ratio (OR) 4.3, 95% CI 1.22-14.43, (p = 0.01)) and overall mortality (OR 4.9, 95% CI 1.68-14.09, (p = 0.001)). We were able to show that preoperative anisocytosis is associated with greater long-term mortality after emergency laparotomy. Increasing RDW demonstrates a relationship with frailty. Given that RDW is readily available at no additional cost, future studies should prospectively validate the role of RDW in the NELA cohort nationally.

Introduction

Every year, approximately 24,000 emergency laparotomies are performed across England and Wales. Postoperative mortality remains high, especially in older patients with comorbidities [1]. Determining surgical risk accurately for individual patients is essential and increasingly emphasised yet remains challenging.

Numerous models have been developed to guide decision making and allow comparison of surgical outcomes following emergency laparotomy. In the United Kingdom, the Portsmouth Physiological and Operative Severity Score for the enUmeration of Mortality and Morbidity (P-POSSUM) model, the Surgical Outcome Risk Tool (SORT) and the National Emergency Laparotomy Audit (NELA) risk model are particularly popular [2].

Despite widespread use, risk prediction tools often have substantial limitations, including resource intensive calculations, dependence on postoperative data and validation bias [2, 3]. Consequently, an ongoing interest remains in identifying new predictors as well as developing more accurate risk prognostication models.

Recent research shows that both frailty and red cell distribution width (RDW) are significant variables in the perioperative setting [4, 5]. To date, neither have been routinely incorporated into surgical risk assessment tools.

The exact link between an elevated RDW and mortality remains unclear but is thought to denote bone marrow dysfunction, systemic inflammation and oxidative stress. Inflammatory pathways mediated by cytokines such as TNF-α and IL-6 inhibit erythropoietin-induced red blood cell maturation and may offer one possible explanation [6].

Importantly, emerging data suggest a strong correlation between anisocytosis, which is reported quantitatively as RDW, and mortality in the older population, perhaps reflecting the multiple physiological impairments related to ageing and frailty [7]. Therefore, RDW may serve as a marker of prior frailty and consequent mortality risk following emergency bowel surgery. Given the availability and the routine reporting of RDW as part of the full blood count, understanding its prognostic value could be both cost-effective and useful for surgical risk stratification in emergency laparotomy patients.

Using our institution’s NELA dataset, we set out to answer three specific questions. First, we examined whether pre-operative RDW values on average are different between emergency laparotomy survivors and non-survivors. Second, we investigated if RDW is a useful predictor of mortality in emergency laparotomy patients and its potential additive value to the NELA model. Finally, we sought to explore whether RDW is independently associated with frailty in this population.

Methods

Data source, patients and outcome measures

This study was a retrospective, single-centre, clinical database analysis conducted at a tertiary London university hospital. Ethical approval for this study was agreed prospectively by the Imperial College London and Imperial College Healthcare NHS Trust Joint Research Compliance Office as well as the Health Research Authority (institutional reference number: 18SM4441/IRAS ID: 242302; HRA: 18/HRA/1860). Under prevailing United Kingdom regulations, individual patient consent was not required given the retrospective nature of the study. Data were pseudo anonymised using the unique NELA identifier. Handling of online NELA data entered by individual NHS trusts adheres to strict information governance standards, which are laid out on the NELA website [8]. All additional administrative or clinical data required were collected as part of routine clinical care and analysed in accordance with the General Data Protection Regulation.

We reviewed the St Mary’s Imperial College Healthcare NHS Trust online NELA database for patients aged eighteen or older who underwent emergency laparotomy between 1st January 2014 and 31st January 2018. The follow-up period ended three months after the data accrual period on 30th April 2018. Our inclusion criteria mirrored those published by NELA [9]. Only the outcome of the index surgery was evaluated if a patient underwent multiple emergency laparotomies during their admission. Patients with no documented operative indication, date of procedure or full blood count were excluded.

Outcomes were 30-day mortality, overall mortality during the follow-up period and frailty. We defined thirty-day mortality as death occurring within 30 days of the index operation. Overall mortality was taken to mean mortality status on 30th of April 2018.

Pre-operative frailty was pragmatically evaluated. We examined the admission clerking of all patients for a documented assessment of frailty using any validated frailty measurement tool. In the absence of such an assessment, the history recorded in the admission clerking was reviewed and compared against the Clinical Frailty Scale (CFS) [10]. Scores greater than or equal to five were taken as frail, which has been shown in the literature to be associated with increased complications as well as mortality [11].

Patient and public involvement

Patients and the public were not involved in the study.

Data collection, missing values and predictor selection

Clinical measurements, comorbidities and expected operative findings were recorded pre-operatively. ASA grade (American Society of Anaesthesiologist physical status classification system) and operative urgency according to the National Confidential Enquiry into Patient Outcome and Death were also included. We classified operative severity according to NELA as major or major+, reflecting surgical immediacy, post-operative length of stay or associated mortality [1, 2].

Blood tests were carried out pre-operatively in our institution’s laboratory and comprised haemoglobin, RDW, white blood cell count, creatinine, urea, sodium and potassium. Full blood counts were measured using the Abbot Alinity-HQ (Abbott, IL, USA) analyser. Creatinine, urea and electrolytes were determined using the Abbott Architect c8000 system (Abbott, IL, USA).

To avoid confounding interventions such as blood transfusions, which could alter the RDW, we understood pre-operative to mean the first set of blood results on admission and not immediately pre-surgery as recorded by NELA. Rarely did in-patients admitted for non-general surgical reasons need an emergency laparotomy. For this small cohort, laboratory values twenty-four hours before surgery were used. Missing NELA database values were cross-referenced with the institution’s clinical information system generating a complete pre-operative dataset.

Candidate risk factors for our mortality analyses were selected a priori based on availability, previous reviews of existing prediction models, national guidelines and research team consensus [2, 12–15]. Thus, the following variables were included: RDW, NELA risk prediction score, haemoglobin, creatinine, and indication for surgery. The NELA risk score incorporates routinely collected predictors (patient demographics, physiological data, laboratory values, and operative details) and has been published elsewhere [2]. A full overview of the included variables can be found in the S1 Annex. Our frailty model contained the covariates sex, age, RDW and haemoglobin.

Statistical methods and model development

We examined baseline patient characteristics across RDW quartiles and checked continuous variables for normality by plotting the data as well as using the Shapiro-Wilk test. Analysis of continuous, non-parametric data was performed using the Wilcoxon-Mann-Whitney test or the Kruskal-Wallis test as appropriate. For categorical variables, the χ2 test or Fisher’s exact test were used to compare frequencies. Associations with P values <0.05 were considered statistically significant.

Using RDW as a continuous variable, we went on to evaluate the prognostic value of RDW at predicting mortality outcomes. Thus, we built two separate nested multivariable Cox regression models (30-day mortality and overall mortality) using the established predictors. Comparing the reduced model (without RDW) with the full model (with RDW) using the likelihood ratio χ2 test allowed us to determine the added predictive value of RDW. Furthermore, the relative importance of RDW in the models was established using an analysis of the variance, allowing for interactions and non-linear effects.

In developing our survival models, it was necessary to combine some of the operative categories with too few patients. We regrouped the variables ‘Colitis’ and ‘Ischaemia’ with the variable ‘Other’. All continuous risk factors had outliers at one end of their distribution. Therefore, the distributions were winsorised at the 5th or 95th percentile as required (see Table 1 of the S1 Annex). Continuous variables were assessed for non-linearity and transformed accordingly. Moreover, several clinically plausible interactions were considered and included if found to be statistically significant. We also checked both models for the proportional hazards assumption and examined for multicollinearity as well as influential observations.

Internal validation of the models was performed using bootstrap resampling, allowing us to estimate the amount of overfitting. The least absolute shrinkage and selection operator (LASSO) method was then employed to shrink regression coefficients. The updated LASSO models enabled us to draw hazard ratio charts presenting point and interval estimates of predictor effects as well as nomograms.

Finally, to investigate the association between RDW as a continuous variable and frailty, we developed a binary logistic regression model. Here we considered RDW as if it were a new diagnostic marker, aiming to characterise its relationship with frailty. We defined frailty to be dichotomous (frail or not frail) and adjusted our model for sex, age as well as haemoglobin. Using approaches similar to the ones outlined above, we checked the underlying model assumptions and penalised our regression analysis for overfitting. Missing data were examined for patterns of missing values and a complete case analysis was carried out.

Publications by Harrell, Spiegelhalter, Pavlou and Torisson informed all modelling algorithms. In designing our models, we adhered to the TRIPOD reporting guidelines [16-20]. A detailed account of their development can be found in the S1 Annex. All statistical analysis was carried out using R v3.5.2 (R Foundation for Statistical Computing, Vienna, Austria, https://www.R-project.org/) and the full code is published on GitHub(https://www.github.com/U601648/RDW_mortality_project).

Results

Overall, 372 emergency laparotomies were recorded during the study period. Sixteen operations were excluded from the final analysis (Fig 1). Baseline participant characteristics are shown by quartiles of RDW in Table 1. For most patients, the laboratory values on admission were used, therefore minimising iatrogenic confounding. However, for fourteen in-patients (3.9%) requiring a laparotomy unrelated to their initial admission, blood tests twenty-four hours before surgery were utilised.

Fig 1

CONSORT diagram of patient enrolment.

Table 1

Baseline characteristics of patients undergoing emergency laparotomy across red cell distribution width (RDW) quartiles.

RDW quartiles
Variable	1st quartile	2nd quartile	3rd quartile	4th quartile
	RDW1 ≥11.7% & <13.1% (n = 93)	RDW2 ≥13.1% & <13.9% (n = 90)	RDW ≥13.9% & <15.1% (n = 86)	RDW≥ 15.1% & ≤27.3% (n = 87)	P value
Demographic
Age	50.0 (37–66)	59.5 (44–76.5)	66.0 (51–77)	64.0 (47–74.5)	<0.001
Female sex (%)	40 (43.0)	48 (53.3)	44 (51.2)	52 (59.8)	0.159
Surgical					<0.001
Obstruction (%)	45 (48.4)	43 (47.7)	43 (50.0)	43 (49.4)
Sepsis (%)	35 (37.6)	36 (40.0)	32 (37.2)	35 (40.2)
Ischaemia (%)	7 (7.8)	5 (5.5)	3 (3.5)	5 (5.7)
Haemorrhage (%)	2 (2.2)	2 (2.2)	3 (3.5)	4 (4.6)
Colitis (%)	3 (3.2)	-	2 (2.3)	-
Other (%)	1 (1.1)	4 (4.4)	3 (3.5)	-
Pre-operative
Median NELA 30-day predicted mortality %	1.2 (0.5–4.9)	2 (0.5–9.3)	4.7 (1.1–11.9)	4.1 (1.75–14.0)	<0.001
ASA score					0.002
ASA 1 (%)	17 (18.3)	17 (18.9)	8 (9.3)	10 (11.5)
ASA 2 (%)	45 (48.4)	34 (37.8)	25 (29.1)	19 (21.8)
ASA 3 (%)	22 (23.7)	23 (25.6)	31 (36.0)	32 (36.8)
ASA 4 (%)	7 (7.5)	14 (15.6)	21 (24.4)	24 (27.6)
ASA 5 (%)	2 (2.2)	2 (2.2)	1 (1.2)	2 (2.3)
Urgency of surgery					0.123
Expedited >18 hours (%)	12 (12.9)	12 (13.3)	23 (26.7)	19 (21.8)
Urgent 6–18 hours (%)	39 (41.9)	31 (34.4)	31 (36.0)	31 (35.6)
Urgent 2–6 hours (%)	36 (38.7)	42 (46.7)	24 (27.9)	34 (39.1)
Immediate <2 hours (%)	6 (6.5)	5 (5.9)	8 (9.3)	3 (3.4)
ECG					0.601
No abnormalities (%)	85 (91.4)	76 (84.4)	78 (90.7)	78 (89.7)
AF rate 60–90 min-1 (%)	2 (2.2)	6 (6.7)	4 (4.7)	2 (2.3)
AF rate >90 min-1 or any other abnormal rhythm, ST changes (%)	6 (6.5)	8 (8.9)	4 (4.7)	7 (8.0)
Cardiac signs					0.826
No failure (%)	80 (86.0)	71 (78.9)	72 (83.7)	67 (77.0)
Diuretic, digoxin, antianginal or hypertensive therapy (%)	10 (10.8)	14 (15.6)	11 (12.8)	17 (19.5)
Peripheral oedema, warfarin therapy (%)	2 (2.2)	2 (2.2)	1 (1.2)	2 (2.3)
Raised JVP or CXR signs (%)	1 (1.1)	3 (3.3)	2 (2.3)	1 (1.1)
Respiratory history					0.611
No dyspnoea (%)	77 (82.8)	72 (80.0)	67 (77.9)	70 (80.5)
Dyspnoea on exertion (%)	11 (11.8)	9 (10.0)	16 (18.6)	10 (11.5)
Dyspnoea limiting exertion (%)	3 (3.2)	6 (6.7)	2 (2.3)	6 (6.9)
Dyspnoea at rest (%)	2 (2.2)	3 (3.3)	1 (1.2)	1 (1.1)
Clinical values
Haemoglobin (gl-1)	143 (133–151)	139 (125–148)	125 (113–139)	120 (96–132)	<0.001
Creatinine (μmoll-1)	76 (67–92)	73(64–101.8)	76 (65–102)	79 (65.5–113)	0.828
Urea (mmoll-1)	5.5 (4.4–7.5)	5.8 (3.6–9.0)	6.1 (4.2–9.0)	6 (4.1–9.35)	0.876
Sodium (mmoll-1)	138 (135–139)	139 (136–141)	138 (135–139)	137 (135–140)	0.059
WBC (x109l-1)	12.2 (8.9–17.3)	10.4 (7.3–13.4)	9.9 (8.0–14.2)	9.9 (6.0–13.8)	0.051
Systolic blood pressure (mmHg)	129 (113–140)	122 (109–138)	124 (107–134)	122 (108–134)	0.484
Pulse (beats min-1)	86 (75–101)	88 (75–102)	84 (76–95)	88 (80–108)	0.204
Perioperative
Operative severity					0.922
Major (%)	60 (64.5)	56 (62.2)	58 (67.4)	56 (64.4)
Major+ (%)	33 (35.5)	34 (37.8)	28 (32.6)	31 (35.6)
Peritoneal soiling					0.826
None (%)	44 (47.3)	44 (48.9)	46 (53.5)	33 (37.9)
Serous fluid (%)	20 (21.5)	17 (18.9)	24 (27.9)	20 (23.0)
Localised pus (%)	5 (5.4)	4 (4.4)	4 (4.7)	6 (6.9)
Free bowel content, pus, or blood (%)	24 (25.8)	25 (27.8)	12 (14.0)	28 (32.2)
Intraoperative blood loss					0.812
<100ml (%)	32 (34.4)	29 (32.2)	26 (30.2)	35 (40.2)
101-500ml (%)	54 (58.1)	56 (62.2)	54 (62.8)	47 (54.0)
501-999ml (%)	5 (5.4)	3 (3.3)	5 (5.8)	4 (4.6)
>1000ml (%)	2 (2.2)	2 (2.2)	1 (1.2)	1 (1.1)
Severity of malignancy					0.826
None (%)	85 (91.4)	77 (85.6)	71 (82.6)	67 (77.0)
Primary only (%)	1 (1.1)	5 (5.6)	10 (11.6)	11 (12.6)
Nodal metastases (%)	1 (1.1)	0 (0)	4 (4.7)	2 (2.3)
Distant metastases (%)	6 (6.5)	8 (8.9)	1 (1.2)	7 (8.0)
Observed 30-day mortality (%)	4 (4.3)	6 (6.7)	10 (11.6)	13 (14.9)	0.061
Observed overall mortality (%)	8 (8.6)	12 (13.3)	20 (23.3)	32 (36.8)	<0.001

Continuous variables are shown as median and interquartile ranges. Categorical variables are shown as a frequency (%). Non-winsorised values were used to draw up the table. P values were calculated using the Kruskal-Wallis test for continuous variables and χ2 test/Fisher’s exact test was used for categorical data (testing for overall difference in RDW quartiles). Obstruction (= small & large bowel obstruction), sepsis (= peritonitis, abdominal abscess, perforation, anastomotic leak), ischaemia (= small & large bowel ischaemia), other (= abdominal compartment syndrome, swallowed foreign body, wound dehiscence, seroma). AF: atrial fibrillation, ASA: American Society of Anaesthesiologist physical status classification system, CXR: chest radiograph, ECG: electrocardiogram, JVP: jugular venous pulse, Major+: all colonic resections, gastrectomy, laparostomy, intestinal bypass, reoperations for bleeding/sepsis, Major: all other including stoma formation, small bowel resection, adhesiolysis, repair of perforated/bleeding ulcer, NELA: National Emergency Laparotomy Audit, RDW: red cell distribution width, WBC: white blood cell count.

All-cause 30-day mortality was 9.3% (n = 33), while overall mortality rose to 20.2% (n = 72) after emergency bowel surgery for the total follow-up period. In those patients who died at 30-days compared to those who survived median RDW levels were consistently higher, 14.9% (IQR 13.7%-16.1%) and 13.8% (IQR 13.1%-15%) respectively, P = 0.007. Median RDW for overall mortality was 13.7% (IQR 13%-14.7%) in survivors versus 14.9% (IQR 13.9%-15.9%) in non-survivors, P<0.001. The cumulative mortality rate increased across RDW quartiles for both follow-up periods and is displayed in Fig 2.

Fig 2

Cumulative mortality rate plots for 30-day and overall mortality post emergency laparotomy by RDW quartiles.

The log-rank test was significant for the total follow-up period χ2 (log-rank) = 25.5, d.f. = 3, p<0.001 (d.f. degrees of freedom). For 30-day mortality the survival lines cross and the log-rank test is unlikely to detect a difference and should not be used for methodological reasons [21].

At 30-days RDW was the least significant predictor. The relative importance of RDW improved considerably for the longer-term mortality model (Table 3 of the S1 Annex). RDW added prognostic value only for the total follow-up period with a calculated percentage of new information of 14%. The overall mortality Cox regression model was internally validated via bootstrapping (1000 resamples) to penalise for possible overfitting. The likely future predictive discrimination measured by Somers’ Dxy rank correlation is 0.46 for the base model and 0.50 for the full model. Optimism adjusted C-statistics were 0.73 and 0.75, respectively. The estimated slope shrinkage was 0.82 and 0.83, suggesting that approximately 0.18/0.17 of the model fitting is noise, especially with regard to calibration accuracy implying moderate overfitting. The calibration curve for the full model is shown in the Fig 6 of the S1 Annex.

LASSO regression was used to shrink regression coefficients and revise the full overall mortality model (mean shrinkage 1.02). To present point and interval estimates of the updated predictor effects a hazard ratio chart was plotted alongside a nomogram for predicting death in emergency laparotomy patients over the total follow-up period (Fig 3).

Fig 3

Hazard ratio chart and nomogram for overall mortality post emergency laparotomy.

Top panel: Estimated hazard ratios (HR) and 95% confidence bars for the overall mortality model. For the NELA risk score interquartile range HR are used, for all other continuous predictors median values are compared to the 90th (RDW, creatinine) or 5th centile (haemoglobin). For example, when RDW changes from its median value (13.9%) to the 90th centile (17.4%), the hazard ratio more than doubles (HR 2.3, 95% CI 1.5–3.5). Standard HRs are presented for surgical indication. Here the hazard ratio is a conventional comparison of the hazard between two groups. Bottom panel: Nomogram for predicting all-cause mortality following emergency laparotomy for the total follow-up period. For each predictor, determine the points assigned on the 0–100 scale and add those points. Plot the result on the Total Points scale and then read the corresponding predictions below it. The linear predictor of a Cox model is a weighted sum of the variables in the models, where the weights are the regression coefficients. Note the effect of interactions, the risk of creatinine is influenced by haemoglobin and the NELA risk score. To illustrate this the 5th and 90th centile was chosen for haemoglobin and the interquartile range for the NELA risk score. RDW: Red cell distribution width (%), hb: haemoglobin (gl-1), cr: creatinine (μmoll-1), nela_risk: NELA risk score, indc_class: indication for laparotomy.

Assessment of frailty was often not recorded, making any judgement about frailty problematic. Hence, it was only possible to draw valid conclusions regarding frailty in 140 (39.3%) patients. Of these, 26 (18.5%) had a formal assessment of frailty documented. All other frailty data (114, 81.5%) were gathered from patient records. Baseline descriptive statistics for the cohort are presented in Table 6 of the S1 Annex.

A significantly higher proportion of patients that died at 30-days were frail (Odds ratio (OR) 4.3, 95% CI 1.22–14.53, P = 0.01). Similarly, the risk of frailty was higher amongst patients that died overall (OR 4.9, 95% CI 1.68–14.09, P = 0.001). Comparing the cohort across groups of RDW demonstrated a higher proportion of frail individuals in each progressive quartile (RDW1: 2 (n = 39), RDW2: 3 (n = 37), RDW3: 7 (n = 32), RDW4: 12 (n = 32)) and was statistically significant, χ2(3, N = 140) = 15.9, p = 0.001.

Based on binary logistic regression analysis, pre-operative RDW was independently associated with frailty in our emergency laparotomy cohort. Validating our model using 400 bootstrap replications the bias-corrected estimate of predictive discrimination was Dxy = 0.462 (C-static 0.73). The corrected Brier score was 0.134, and the estimated maximum calibration error in predicting frailty was 0.06 (Table 8 of the S1 Annex). We depicted the fitted model by computing odds ratios with their respective 95% confidence intervals and graphed the association of RDW with frailty in NELA patients, estimated for a range of different ages (Fig 4).

Fig 4

Frailty logistic regression model.

The left-hand panel displays an estimated odds ratio (OR) chart and respective 95% confidence intervals. For example, when RDW changes from the 50th to the 90th percentile (13.8% to 17.3%) the odds ratio of being frail is 2.9 (95% CI 1.4–6.4). The odds for age (OR 1.8, 95% CI 1–3.4) are for the 25th and 75th percentile, while for haemoglobin (OR 0.8, 95% CI 0.4–1.5) they are based on the 10th and 50th percentile. The right-hand panel illustrates the effect of RDW on the probability of frailty for emergency laparotomy patients, estimated for different ages. The age cut-offs represent the 10th, 25th, 50th, 75th and 90th percentile (n = 140). RDW: red cell distribution width (%), age_at_adm: age at admission (years), hb: haemoglobin (gl-1).

Discussion

To our knowledge, this study is the first to examine pre-operative RDW and mortality, its potential added predictive value, and its relationship with frailty in emergency laparotomy patients. We found that RDW values, on average, were higher in non-survivors. Moreover, there was a distinct gradient in overall mortality risk associated with increasing RDW. This association remained after accounting for the NELA risk score, haemoglobin, creatinine and operative indication for overall mortality but not shorter-term 30-day mortality.

In the peri-operative setting, anisocytosis has been mainly associated with long-term mortality after surgery [22]. However, more recently, Abdullah and colleagues described a link between 30-day mortality and pre-operative RDW in patients 65 years or older undergoing noncardiac surgery [4]. This differs from our findings and others, where RDW was not a convincing predictor of death at 30-days [23, 24]. In line with a recently published retrospective database study, RDW had a stronger association with overall mortality in our emergency laparotomy cohort [24]. A discrepancy, which is likely to have arisen due to differences in study population and methodology. For example, we did not dichotomise RDW using sensitivity analyses but explored RDW as a quantitative variable avoiding the categorisation of an inherently continuous marker. Furthermore, the choice of regression coefficients is likely to account for much of the observed disparity. Our findings show that the composite NELA risk score is the main predictor of all-cause mortality in both models. The NELA tool was developed to produce risk-adjusted 30-day postoperative mortality rates [2]. Thus, RDW is probably not influential enough in our model at 30-days, lacking in discriminatory power and adding little in predictive value compared with the NELA risk score. Conversely, the NELA risk score was not designed with long-term mortality in mind and may explain the improved prognostic influence of RDW on our overall model.

Though we did not expressly investigate how prognostic factors impact outcome over time, it is biologically plausible that markers differ in their predictive ability in a time-dependent manner. The NELA model primarily reflects perioperative events, which may have less influence on patients who survive long-term, usually because of treatment with curative intent. Hence, the predictive worth of the NELA risk score is likely to decrease with time after laparotomy. In contrast, pre-operative anisocytosis may indicate chronically reduced physiological reserve, making it possibly a better indicator of longer-term mortality [25]. An interesting follow-up study would be to formally evaluate at what time point RDW, as a prognostic marker, has the most significant impact on mortality prediction.

Although numerous studies, including ours, have shown an association between higher RDW and decreased survival, the exact causal relationships remain elusive and are likely to be multifactorial [4, 24, 25]. Various hypotheses have been suggested, all of which involve systemic factors that alter erythrocyte physiology, such as oxidative stress, inflammation, malnutrition and telomere length [25, 26]. Nonetheless, there is an emerging consensus that anisocytosis reflects profound physiological dysregulation.

While many of the above mechanisms are likely to be similar to those implicated in the pathophysiology of anaemia, we found anisocytosis to be independent of haemoglobin concentration. This is in keeping with findings published by Patel and colleagues [26]. Equally, haemoglobin concentration was not a meaningful predictor of all-cause mortality in our study. A similar conclusion was reached during the development of the NELA risk prediction tool, leading to its exclusion from the model [2].

RDW is also strongly associated with advancing age and a higher disease burden [27]. More recently, a connection between RDW and frailty has been suggested, an association that we were able to support in our explorative analysis [27]. Intriguingly, frailty and anisocytosis appear to share similarities in their proposed pathophysiological mechanism [25, 28]. Thus, RDW is a possible integrative biomarker reflecting the multiple biological impairments related to increasing frailty and indirectly ageing, perhaps thereby explaining its additional predictive worth.

We acknowledge several limitations, including the single-centre, retrospective observational design of the study and its relatively small sample size, restricting its overall generalisability. While national inclusion criteria mitigate selection bias, our findings ideally require prospective confirmation across the whole NELA cohort. At present, the NELA project does not routinely collect RDW. Since RDW is easily measured as part of the full blood count, including it prospectively in large nationally or internationally collected datasets may validate its effectiveness and offer valuable insights prognostically.

A further shortcoming is that we did not account for blood transfusions, which could modify RDW. We used admission blood tests to attenuate the confounding risk of perioperative blood transfusions, but a small proportion of patients underwent laparotomy as in-patients. In an attempt to adjust for additional risk factors, we applied the amalgamated NELA score. Nevertheless, we cannot exclude the possibility of residual confounding. In particular, we did not account for nutritional deficiencies (folate, cobalamin, iron) and cancer, similar to many studies on RDW. However, a large community-based study in the United States examining RDW in middle-aged and older adults found RDW to predict mortality independent of these confounding factors [7].

Moreover, we developed our frailty model, excluding a large number of patients with missing data. The distribution of variables across risk factors was similar in patients with complete and missing frailty outcomes, suggesting that the data were missing at random (Table 7 of the S1 Annex). Reassuringly, the prevalence of frailty in our cohort mirrored a national multicentre study specifically examining frailty in NELA patients [29]. In the majority of patients, frailty was established using the clinical notes. Admittedly subjective, the simplicity of the CFS score facilitates this and is thought to be appropriate in the literature [11].

Lastly, we recognise that internal validation demonstrated overfitting for both our models. This is most likely due to the high number of parameters, including screening for non-linear terms and global interaction tests. However, our models were exploratory and not meant to be new parsimonious prediction tools. Thus, we emphasised the inclusion of clinically relevant variables alongside interactions/non-linear terms in the trade-off with overfitting [19].

Our study also had various strengths, specifically minimal loss of predictor values, a priori choice of covariates and a robust approach to model development. We used advanced methods to address non-linearity, interactions, internal validation and presented our models graphically with these complexities in mind. Importantly, we avoided the categorisation of RDW and many of its associated problems [11]. Some of these include the heterogeneity of diagnostic and prognostic cut-offs in the literature and unmet standardisation of erythrocyte sizing [7]. Crucially, specific cut off values imply that the relationship with an outcome is flat on either side of the chosen value, which biologically is seldom plausible [30]. Indeed, we were able to demonstrate that mortality increases across what is considered the normal range of RDW, representing a continuum of risk and is depicted in our nomogram (Fig 4).

A key strength of investigating RDW is its availability at no additional cost since it is routinely performed as part of the full blood count. Similarly, use of the CFS to screen for frailty is straightforward and uses readily available clinical information. While concerns around its applicability in patients below 65 years of age exist, it has been used successfully in adult emergency surgical admissions regardless of age [11].

Despite mounting evidence that anisocytosis is associated with increased long-term mortality following surgery, large-scale prospective studies are now needed to validate its predictive utility [4, 24]. Going forward, investigators should focus on RDW as a continuous variable to develop valid prediction models rather than classification tools based on subjective thresholds. Moreover, these studies should now assess the added predictive value of RDW to determine if pre-operative anisocytosis enhances current risk-stratification tools. In turn, superior risk prediction tools could allow more meaningful informed consent and shared decision making between patients and healthcare professionals.

At present, it remains unknown whether RDW is a modifiable risk factor perioperatively, including the elective setting. It would be interesting to establish if targeting factors reflected in the RDW improves surgical outcomes. Should tailored interventions such as physical rehabilitation, nutritional support or immunomodulation prove beneficial, this would further strengthen the argument to use RDW to identify individual patients at risk [24].

Conversely, the idea that frailty contributes to increased mortality following emergency surgery is not new, nor is the concept of integrating frailty into surgical risk assessment [5]. However, whether increased RDW, as a measure of biological vulnerability, offers a valid link with frailty should now be formally investigated.

Finally, pre-operative risk models for emergency laparotomies are based on retrospective database analyses of patients undergoing surgery [2]. We know little about patients who met the criteria for surgery but did not proceed due to personal choice or perceived high risk [31]. Future research must establish the predictive value of RDW for all patients with or without surgical intervention to understand its pre-operative worth fully.

Conclusions

We established that anisocytosis as reflected in the RDW value is associated with higher rates of postoperative mortality following emergency laparotomy. Furthermore, our analysis tentatively supports the notion that increased RDW is a possible marker of physiological dysregulation relevant to frailty [24]. While further research is needed to understand these mechanisms fully, RDW seemingly provides prognostic information that could inform future risk prediction tools. Accordingly, we explored how to quantify the added prognostic value of RDW without resorting to categorisation. Although oversimplified for illustration, our models demonstrated a statistically efficient way to investigate the relative merit of RDW. However, whether adding RDW as a global marker of homeostasis to surgical prognostication tools will improve patient management and outcome remains to be seen.

(DOCX)

Click here for additional data file.

Statistical discussion.

(DOCX)

Click here for additional data file.

14 Feb 2022

7 Mar 2022

Journal requirements:

PLOS ONE style requirement:

Please ensure that your manuscript meets PLOS ONE’s style requirements, including those for file naming.

The style and file naming have been amended as required.

2. Ethics statement:

In the ethics statement in the manuscript and in the online submission form, please provide additional information about the patient records/ samples used in your retrospective study. Specifically, please ensure that you have discussed whether all data/samples were fully anonymised before you accessed them and/or whether the IRB or ethics committee waived the requirement for informed consent. If patients provided informed written consent to have data/samples from their medical records used in research, please include this information.

Many thanks for this. We have amended the ethics statement as suggested. The added sentence reads: “Under prevailing UK regulations, individual patient consent was not required given the retrospective nature of the. Data were pseudo anonymised using the unique NELA identifier. Handling of online NELA data entered by individual NHS trusts adheres to strict information governance standards, which are laid out on the NELA website [8]. All additional administrative or clinical data required were collected as part of routine clinical care and analysed in accordance with the General Data Protection Regulation.”

3. Competing interests:

Thank you for stating the following in the Competing Interests section: “MB was a Health Service Research Centre fellow with the NELA project form August 2018 – August 2019. All other authors (JLG, REB, SJB) declare that they have no conflict 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.

Many thanks for highlighting the above. Our revised statement now includes the sentence as suggested.

MB was a Health Service Research Centre fellow with the National Emergency Laparotomy from August 2018 to August 2019. All other authors (JLG, REB, SJB) declare that they have no conflict of interest. We received no financial support for the research, authorship or publication of this article other than Imperial Biomedical Research Centre infrastructure support for Professor Brett. This does not alter our adherence to PLOS ONE policies on sharing data and materials.

4. Direct Billing:

Please note that in order to use the direct billing option the corresponding author must be affiliated with the chosen institute. Please either amend your manuscript to change the affiliation or corresponding author or email us at plosone@plos.org with a request to remove this option.

Thank you for pointing this out, we have amended the corresponding author as requested. The corresponding author will be Professor S. Brett (stephen.brett@imperial.ac.uk).

5. Reference list review:

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.

Thank you very much for giving us the opportunity to double check our references. We can confirm that none of the articles have been retracted. Furthermore, we have included one additional reference to clarify handling of confidential data in the ethics section (reference number 8) and is marked in red in the reference list.

Reviewer #1:

1. Thank you very much for the opportunity to review this interesting research article. The introduction gives a strong basis for the study and its relevance. My few comments ask for some clarification with regards to the methods.

Thank you very much for your kind words.

2. Page 4, line 97 - There should be some justification for the 3 month follow-up period as opposed to 6 months or one year.

The follow-up period was chosen a priori. We looked at 30-day and overall survival after laparotomy. We reviewed data from 1st January 2014 to 31st January 2018 and included all patients during this period. Patients recruited at the beginning of the study naturally have a longer follow-up period compared to patients at the end of the study. Thus, patients with an emergency laparotomy in early 2014 potentially were followed-up for 4 years. Conversely a patient undergoing emergency bowel surgery in late January would have had a minimum follow-up of three months.

We chose 30-day and overall mortality (with a minimum 3-month follow-up) endpoints for the following reasons. Mortality at 30 days is conventionally accepted to reflect early events. Beyond 30 days death is less likely to reflect periprocedural related mortality. Overall mortality allows the rate of death in this cohort to be examined generally and to pinpoint when it occurs. Examining the cumulative mortality rate plot in Annex 1 (Figure 1) the majority of deaths post emergency laparotomy occur in the first three months.

Although we accept that a minimum 3-month follow-up period may appear short, most deaths tend to happen in this early phase. Moreover, only a small number of patients (n=21 (6%)) did not have a follow-up period of at least six months (patients operated after the 1st November 2017). We hope that this goes some way to explain our choice of follow-up period.

3. Given the NELA risk prediction score is included as part of the model, the components of this should be more explicit for those who are not familiar with this score.

We are grateful to the reviewer for highlighting this perfectly reasonable point. We have added the following sentence to our manuscript:” The NELA risk model incorporates routinely collected predictors (patient demographics, physiological data, laboratory values, and perioperative details) and has been published elsewhere [2]. A full summary of the included variables can be found in Annex 1.” The description of the risk factors can be found under Data collection in Annex 1.

4. There needs to be some justification for making the ASA score binary and not categorical.

Although convenient for tabulation and data presentation we agree that dichotomising ASA for these reasons is not justified, particularly as it leads to significant loss of information and statistical power. Hence, we have included all ASA grades in Table 1 as suggested.

5. Please give a justification for differentiating major and major + in operative severity and how different surgeries were categorized into each of these areas.

We agree that dichotomising operative severity is an oversimplification. The UK National Emergency Laparotomy Audit takes the view that all emergency laparotomies are major procedures, defined as surgery carried out within 24 hours following decision to operate. To nuance a colonic resection from adhesiolysis (division of adhesions causing intestinal obstruction, without resection) the differentiation major and major+ was introduced. This simplified binary categorisation is used across annual NELA reports and in the paper describing the NELA risk model (Eugene et al. Development and internal validation of a novel risk adjustment model for adult patients undergoing emergency laparotomy surgery: The National Emergency Laparotomy Audit risk tool. BJA 2018, 121 (4): 739-748).

In view of the well-made point made by the reviewer we have clarified the binary categorisation by adding the sentence: “Operative severity was classified according to NELA as major or major+, reflecting surgical immediacy, post-operative length of stay and mortality [1,2]” to the data collection, missing values, and predictor selection paragraph. The footnote below table 1 in the main article details how different surgeries were categorised into each group.

6. Why not analyze RDW as a continuous variable instead of in quartiles?

Many thanks for this crucial comment, indeed we should have made this clearer. Patient characteristics were examined across RDW quartiles and presented in Table 1. However, all modelling avoided categorisation and RDW was used as a continuous variable throughout. Annex 1 elaborates on this in depth, but we agree that in the main manuscript this is not made obvious enough until the discussion.

We have amended the sentences describing the models using RDW as a continuous predictor. The sentences are as follows:

“Using RDW as a continuous variable, we went on to evaluate the prognostic value of RDW at predicting mortality outcomes.” [page 6, line 152]

“Finally, to investigate the association between RDW as a continuous variable and frailty, we developed a binary logistic regression model.” [page 7, line 172]

7. Table 1 – it is unclear what the p value of the surgical indication is describing. Please clarify.

Thank you for noting this. We agree that the terminology appears confusing. Surgical indication relates to the surgical pathology requiring a laparotomy. The p value relates to the overall difference in surgical disease across RDW quartiles. We have aligned the diagnostic groupings for greater visual clarity (all are now left justified in the respective cell) and changed surgical indication to surgical pathology. We have detailed the breakdown of surgical pathology in the footnote of Table 1.

8. Figure 3 – I am unsure what this figure is aiming to show. Maybe there is a better way to compare these variables that is more easy to read? For example, the RDW vs HR is explained, but it is not clear to me how to interpret the surgical indication variables. Is NELA score used as a reference point in this case?

Many thanks for this valuable feedback. We agree that the left-hand hazard ratio plot is complex. Nonetheless we feel it serves two important purposes. The first is to visually represent estimated predictor effects. The second is to present both point estimates (for surgical indication) and interval estimates for continuous variables (avoid categorisation). The later demonstrates how hazard ratios change for predictors in the model over a range of values.

We attempted to make this clear in the footnote below the graph but accept that this could be made clearer. Hence we have changed the sentence “Simple HR are presented for categorical predictors” to “Standard HRs are presented for surgical indication. Here the hazard ratio is a conventional comparison of the hazard between two groups.”

9. Could you also comment on why Cr was used as a variable in the right panel nomogram and not the other variables such as NELA score or hemoglobin?

We are grateful for this valuable comment regarding Figure 3. The intention of the nomogram was to depict the model while enabling interactions to be demonstrated. The risk of creatinine is influenced by haemoglobin and the NELA risk score. Thus, using different values for haemoglobin (5th and 90th centile) and NELA risk (25th, 50th and 75th centile) we illustrate how creatinine as a risk predictor is modified. To clarify this, we have changed the footnote below the graph to include the following sentence:” To illustrate this the 5th and 90th centile was chosen for haemoglobin and the interquartile range for the NELA risk score”. We have discussed these interactions in greater depth in Annex 1.

10. The discussion is also very focused and well written.

Thank you for this kind comment.

Reviewer #2:

1. This study demonstrates clearly that prognosis of patients undergoing laparoscopy is more adverse in the case of a larger red cell distribution width. The authors investigated a large cohort of patients and performed diligent statistical analyses. The study deserves interest and contributes to insights in the clinical impact of red cell distribution.

Many thanks for this kind comment.

2. The study would benefit from an outlook for the future in which way the results of this study could become beneficial for patients in the surgical setting: For instance, identifying patients at increased risk of peri- and post-surgery complications who may benefit from a closer monitoring in the in- and outpatient setting.

Thank you for emphasising the clinical aspect and potential benefit to patients in the future. We agree and have added the following paragraph to strengthen this aspect:

“In turn, superior risk prediction tools could allow more meaningful informed consent and shared decision making between patients and healthcare professionals. At present, it remains unknown whether RDW is a modifiable risk factor perioperatively, including the elective setting. It would be interesting to establish if targeting factors reflected in the RDW improves surgical outcomes. Should tailored interventions such as physical rehabilitation, nutritional support or immunomodulation prove beneficial, this would further strengthen the argument to use RDW to identify individual patients at risk [24].”

Submitted filename: Response_to_reviewers.docx

Click here for additional data file.

14 Mar 2022

Exploring red cell distribution width as a potential risk factor in emergency bowel surgery – a retrospective cohort study

PONE-D-21-35723R1

Dear Dr. Brett,

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,

Itamar Ashkenazi

Academic Editor

PLOS ONE

14 Apr 2022

PONE-D-21-35723R1

Exploring red cell distribution width as a potential risk factor in emergency bowel surgery – a retrospective cohort study

Dear Dr. Brett:

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

Dr. Itamar Ashkenazi

Academic Editor

PLOS ONE