Comparing physical assessment with administrative data for detecting pressure ulcers in a large Canadian academic health sciences centre.

OBJECTIVE: This study aimed to compare classification of pressure ulcers from administrative data with a gold standard assessment, specifically; pressure ulcers confirmed by an independent physical assessment performed by trained nurse surveyors.
SETTING: A retrospective analysis of pooled cross-sectional samples of inpatients assessed across 3 consecutive prevalence surveys in a large academic health sciences centre between 2012 and 2013. PARTICIPANTS: There were 2001 patients for whom physical and chart assessments were completed, and for whom a discharge abstract was also available at the time of analysis. The cohort's mean age was 65 years and 55% were women.
RESULTS: Based on the physical assessment findings, 14.6% of patients (n=292) had at least 1 pressure ulcer, with a total of 345 pressure ulcers documented among these patients: (stage I=162; stage II=120; stage III=22; stage IV=22 and unstageable=19). Based on coded information, 78 (3.9%) of patients had a pressure ulcer. Of patients with a pressure ulcer determined by the physical assessment, only 21% also had a pressure ulcer captured in the administrative data. Furthermore, only 6% of the patients with a hospital-acquired pressure ulcer, stage II or greater determined by the physical assessment were coded in the Discharge Abstract Database (DAD).
CONCLUSIONS: The results of this study demonstrate that coding in the DAD may under-report and fail to accurately reflect the true burden of pressure ulcers in hospitalised patients. This may occur because the presence of pressure ulcers is currently documented in the health record by nurses and not by physicians, yet the administrative data recorded in the DAD only includes physician documented pressure ulcers. We recommend enhancements to the coding methods to monitor and report on pressure ulcers. Published by the BMJ Publishing Group Limited. For permission to use (where not already granted under a licence) please go to http://www.bmj.com/company/products-services/rights-and-licensing/.

We studied a large number of patients in a large academic health sciences centre using the physical assessment as the gold standard comparator. There are no other similar studies of this important problem.

We compared any pressure ulcer observed on prevalence day to any pressure ulcer documented in the administrative data.

We compared two different types of prevalence estimates: cross-sectional and period prevalence. Owing to the nature of each type of prevalence estimate, the sample population may have been different between each group with longer stay patients being over-represented in the cross-sectional prevalence.

Introduction

Pressure ulcers are a significant issue among hospitalised patients worldwide. In the Canadian context, the estimated prevalence ranges from 23.9% to 29.7% among acute care patients.1 Avoiding skin breakdown and the development of pressure ulcers is a recognised priority for ensuring quality of care and potential cost-savings.2 According to the US-based National Pressure Ulcer Advisory Panel (NPUAP), the cost of caring for pressure ulcers has been estimated at over US$11 billion per year. Since pressure ulcers are for the most part preventable, there is a need to identify the at-risk populations and to target early intervention strategies. Despite an increased awareness surrounding the burden of pressure ulcers and the importance of prevention and treatment, there is a lack of accurate population-based methods to detect and monitor pressure ulcer rates. This creates challenges for the measurement of outcomes following efforts to prevent and reduce pressure ulcers over time.

Administrative data are one source for estimating pressure ulcer incidence and prevalence. Each inpatient hospital encounter is summarised on discharge through abstracts submitted to the Discharge Abstract Database (DAD), based on the Coding Standards for the International Classification for Diseases and Related Health Problems, 10th Revision (ICD-10). Each diagnosis recorded in the DAD must be coded according to ICD-10, and assigned a Diagnosis Type Code, representing the influence of the condition on the patient's treatment.3 According to these standards, diagnosis code and diagnosis type are taken directly from the physician's documentation. DAD coding and diagnosis typing may also be sought from the documentation of other healthcare professionals, but only if they have been designated with primary responsibility for the patient's care.3 Further complicating the coding is the fact that the staging system used for coding in the DAD may differ from the NPUAP staging guidelines that are followed by the majority of nurses and other healthcare professionals in Canada.4

Owing to the potential variation in documentation practices between healthcare professionals and the provider specifications in the coding guidelines, we hypothesised that the current estimates of pressure ulcers based on these administrative codes likely do not accurately reflect the true burden of this condition.5 If the information in the DAD does not accurately capture the pressure ulcers, this can have a significant impact on healthcare policy that often relies on statistical information from these types of electronic sources. Therefore, the overall aim of this study was to compare classification of pressure ulcers from the DAD with a gold standard assessment, specifically, pressure ulcers confirmed by an independent physical assessment performed by trained nurse surveyors.

Methods

Study design and setting

This study involved a retrospective analysis of pooled cross-sectional samples of inpatients at a Canadian academic health sciences centre with more than 1127 inpatient beds. We received Research Ethics Board approval to use these data for research purposes.

Participants

All patients who were admitted and assessed on 1 of 3 consecutive prevalence survey days between 2012 and 2013 were included. Each prevalence survey consisted of a 1-day cross-sectional survey where nurse surveyors collected data from inpatients' health records (ie, a chart assessment) and performed a physical assessment of all inpatients. The content of the survey was selected to identify patients at risk of developing pressure ulcers, to determine the presence of pressure ulcers, and to assess health providers' adherence to the hospital's policies and programmes. The number of pressure ulcers, their location and their severity ‘stage’ (according to the NPUAP definitions that were in effect during the study periods) were recorded for each patient. These were further classified as ‘developed in hospital’ or ‘present on admission’, based on the results of the chart assessment. All surveyors received standard training on staging pressure ulcers. On the survey day enterostomal therapy nurses, specialists in wound care, were also available to help with staging pressure ulcers if needed. There has been similar prevalence surveys conducted since 1993 to provide information on the presence and severity of pressure ulcers, among other conditions. For purposes of this analysis, we used the data from 2012 to 2013 because the data from these 2 years were available electronically.6

The case selection process is described in figure 1. We pooled data from patients who were present on 1 of the 3 consecutive prevalence days (25 April 2012, 21 November 2012 and 25 April 2013). A small number of patients did not undergo a physical or chart assessment for various reasons (eg, patient was receiving a procedure at the time of the prevalence survey or the patient's chart was not available at the time of survey). We excluded these patients. If a patient was in the hospital on more than one prevalence survey day, or if a patient's discharge abstract was not available at the time of analysis (ie, where a patient was still in hospital, or the discharge abstract was not complete) they were also excluded. Each of the included patient encounters was also summarised in an abstract that was submitted to the DAD and that included a diagnostic code (ICD-10 classification) and a diagnostic type (eg, M: most responsible diagnosis; type 1: preadmission comorbidity; type 2: postadmission comorbidity).3

Figure 1

Derivation of study cohort.

Analysis

To assess the accuracy of administrative data in identifying pressure ulcers among patients discharged from hospital, we compared pressure ulcer cases identified in the DAD against ‘nurse-confirmed’ pressure ulcers that were assessed and documented on 1 of the 3 prevalence days. The analysis was divided into three parts. First, we compared patients with any pressure ulcer identified on prevalence day to any pressure ulcer coded in the corresponding DAD records (ICD-10 codes: L89.0–L89.9). Second, we narrowed the analyses to patients with a pressure ulcer that developed after admission only, compared with pressure ulcers coded as a type 2 diagnosis in the DAD record (ie, postadmission comorbidity). Third, we further limited the comparison to patients with a stage II pressure ulcer developed in hospital against pressure ulcers coded as type 2 diagnoses in the corresponding DAD records. Stage I pressure ulcers may not be recognised or documented as frequently as the more severe stages of the condition,7 8 and therefore we expected higher agreement between the prevalence data and the DAD when we limited the analyses to stage II pressure ulcers and above.

We calculated the sensitivity, specificity, positive predictive value and negative predictive value for each of the comparisons, using the nurse-confirmed physical assessment as the ‘gold standard’. We also calculated the κ statistic and 95% CIs to account for the level of agreement due to chance, where a κ of 1.0 would indicate perfect agreement and a κ of 0.0 would indicate agreement based on chance alone.9 We described the study population using basic demographic variables and clinical characteristics. We also calculated the Elixhauser Comorbidity Score developed by van Walraven et al10 to summarise the comorbidity in this patient sample.

The data analysis for this study was generated using SAS software, V.9.3, 2011 SAS Institute, SAS and all other SAS Institute product or service names are registered trademarks or trademarks of SAS Institute, Cary, North Carolina, USA.

Results

Across 3 consecutive prevalence days, there were 2001 patients for whom physical and chart assessments were completed, and for whom a discharge abstract was also available at the time of analysis (figure 1). The cohort's mean age was 65 years and 55% were women. The cohort is described further in table 1.

Table 1

Description of the cohort

Total patients	N=2001*
Age (years)
Mean	65.1±18.6
Median	68 (53–80)
Gender
F	1099 (54.9)
M	902 (45.1)
Department
Medicine	897 (44.8)
Surgery	744 (37.2)
Other	122 (6.1)
Obstetrics/gynaecology/newborn care	120 (6.0)
Family practice	118 (5.9)
Elixhauser score
Mean	4.9±6.4
Median	3 (0–8)
Total LOS (days)
Mean	30.1±48.3
Median	15 (6–31)
Acute LOS (days)
Mean	19.7±29.1
Median	12 (6–23)

Continuous characteristics are presented with the mean±SD and median (IQR).

ALC patients are those who no longer require acute care services and are waiting to be discharged to a suitable care setting in the community.

*Categorical characteristics are presented with percentages.

Acute LOS (total LOS)−(days spent in ‘ALC’ status).

ALC, alternate level of care; F, female; LOS, length of stay; M, male.

Based on the physical assessment findings, 14.6% of patients (n=292) had at least 1 pressure ulcer, with a total of 345 pressure ulcers documented among these patients: stage I=162; stage II=120; stage III=22; stage IV=22 and unstageable=19 (ie, some patients had multiple pressure ulcers). In contrast, the results from the coded information showed that only 78 (3.9%) of patients had any pressure ulcer. Table 2 provides the prevalence of pressure ulcers among the prevalence sample and table 3 provides the prevalence of pressure ulcers as coded in the DAD.

Table 2

Pooled data from three consecutive prevalence surveys

Patients with physical assessment	N=2001 (%)
Patients with at least one pressure ulcer	292 (14.6)
Developed while in hospital, any stage	201 (10.0)
Developed while in hospital, stage II or greater	89 (4.4)

Table 3

DAD records corresponding to patients observed on one of three consecutive prevalence surveys

DAD records		N=2001 (%)
ICD-10 code	Patients with at least one code for:
L89.0–L89.9	Pressure ulcer, any stage	78 (3.9)
L89.1–L89.9	Pressure ulcer, stage II or greater	71 (3.5)
Diagnosis type
Type 1	Pressure ulcer developed while in hospital	16 (0.8)
Type 2	Pressure ulcer on admission	45 (2.2)

DAD, Discharge Abstract Database; ICD-10, International Classification for Diseases and Related Health Problems, 10th Revision.

Of the nurse-confirmed cases, only 21% (n=61) were captured in the administrative data; however, only a small percentage (1%) of patients with a pressure ulcer coded in the DAD did not have a pressure ulcer on prevalence day (either not present, not observed or it may have developed later; table 4).

Table 4

Patients with a pressure ulcer present on physical assessment, compared with patients with a pressure ulcer coded in the DAD

	Pressure ulcer on assessment
	Yes	No	Total
Any ICD-10 pressure ulcer code
Yes	61	17	78
No	231	1692	1923
Total	292	1709	2001

Sn=0.21; Sp=0.99; PPV=0.78; NPV=0.88; κ=0.29 (95% CI 0.23 to 0.35).

DAD, Discharge Abstract Database; ICD-10, International Classification for Diseases and Related Health Problems, 10th Revision; NPV, negative predictive value; PPV, positive predictive value; Sn, sensitivity; Sp, specificity.

The agreement between the two sources decreased when we analysed the sample by patients who developed a pressure ulcer after admission compared with patients with a type 2 diagnosis code in the DAD (postadmission comorbidity; table 5). Only 4% (n=8) of the patients with a pressure ulcer that developed in hospital were also coded in DAD as a type 2 diagnosis.

Table 5

Patients who developed a pressure ulcer while in hospital, compared with patients with a type 2 pressure ulcer coded in the DAD

	Pressure ulcer developed while in hospital
	Yes	No	Total
Any type 2 ICD-10 pressure ulcer code
Yes	8	8	16
No	193	1792	1985
Total	201	1800	2001

Sn=0.04; Sp=1.00; PPV=0.50; NPV=0.90; κ=0.06 (95% CI 0.01 to 0.10).

DAD, Discharge Abstract Database; ICD-10, International Classification for Diseases and Related Health Problems, 10th Revision; NPV, negative predictive value; PPV, positive predictive value; Sn, sensitivity; Sp, specificity.

Furthermore, only 7% (n=6) of the patients with at least one pressure ulcer stage II or greater that developed in hospital and was observed through physical assessment were also coded in DAD (table 6). The κ observer agreement calculations identified, at best, ‘fair’ agreement between the two sources (κ=0.29), but otherwise showed only ‘slight’ agreement between the two detection methods, according to the categories presented by Landis and Koch.11

Table 6

Patients who developed a stage II or greater pressure ulcer while in hospital, compared with patients with a type 2 pressure ulcer coded in the DAD

	Pressure ulcer stage II or greater developed while in hospital
	Yes	No	Total
Any type 2 ICD-10 pressure ulcer code
Yes	6	10	16
No	83	1902	1985
Total	89	1912	2001

Sn=0.07; Sp=0.99; PPV=0.38; NPV=0.96; κ=0.10 (95% CI 0.02 to 0.18).

DAD, Discharge Abstract Database; ICD-10, International Classification for Diseases and Related Health Problems, 10th Revision; NPV, negative predictive value; PPV, positive predictive value; Sn, sensitivity; Sp, specificity.

Discussion

The results of this study suggest that a large proportion of pressure ulcers may not be captured in the administrative data that are routinely collected to summarise a patient's stay in hospital. This is consistent with a recent study by Meddings et al5 which found that administrative data were a poor indicator of pressure ulcer performance in a large sample of California hospitals, compared with pressure ulcers detected through surveillance. The inconsistencies identified by the authors highlight why pressure ulcer rates from administrative data are under-reported and are unlikely suitable for performance benchmarking efforts. These shortcomings in the administrative data are, in part, related to poor pressure ulcer documentation practices in the patient record. In Sweden, Gunningberg and Ehrenberg7 found that paper-based pressure ulcer documentation was poor, but that documentation improved with the implementation of an electronic health record.12 In the USA, Dahlstrom et al13 found an opposite effect with respect to the completeness of documentation, when documentation was moved to an electronic system; however, this trend subsequently reversed for nurses, but not among physicians.

A potential explanation for the differences observed between the prevalence survey data and the DAD relates to who was documenting the pressure ulcer. This is largely due to the fact that pressure ulcers are routinely documented by nurses and not by physicians. In the USA, Arora et al14 observed that the rate of pressure ulcer risk assessment among nurses was 100%, compared with just under 3% among physicians, and that physicians documented fewer than one out of every two pressure ulcers identified by nurses. In a related study by Dahlstrom et al13 following a campaign to improve pressure ulcer documentation, researchers observed that nurses documented almost all pressure ulcers (96.7%) compared with significantly fewer documented by physicians (70.6%). This difference increased when the researchers considered ‘complete’ documentation, only (46.2% and 15.2% among nurses and physicians, respectively). Therefore, if nurse documentation is not being considered when pressure ulcers are being coded, then the administrative data are likely missing a large proportion of these cases.

However, in the same study, the authors noted that even after the intervention, fewer than half of the cases documented by nurses contained ‘complete’ information (ie, size, location and stage).13 In Iceland, Thoroddsen et al15 also observed that size and category were documented 11% and 55% of the time, respectively, for recorded pressure ulcers, but found that location was documented more consistently (88%). Similarly, Gunningberg and Ehrenberg7 found that nursing documentation captured pressure ulcer location 96.6% of the time, but only captured size 15% of the time and category was not documented in any of the records. Therefore, although there is evidence that nurses document pressure ulcers more often than physicians, the quality of the documentation, overall, may still be lacking.

A few limitations should be considered when interpreting our study's results. The main limitation is that we compared any pressure ulcer observed on prevalence day to any pressure ulcer documented in the administrative data. Therefore, it is possible that we recorded agreement between the two sources whereas, in reality, the pressure ulcers were mutually exclusive.

A related limitation is that we compared two different types of prevalence estimates: cross-sectional and period prevalence. In general, cross-sectional prevalence studies are limited because they represent a single point in time. When patients are sampled in this way, those who are in hospital longer are more likely to be sampled during the course of their stay than those with a shorter length of stay. Therefore, if a condition being studied is related to a longer length of stay, patients with this condition will be over-represented in a cross-sectional sample.16 Given that pressure ulcers are an example of a condition that can extend a patient's length of stay,17 18 and where a longer length of stay can also increase the risk of developing the condition,2 the prevalence of pressure ulcers measured on each prevalence day is likely to overestimate the prevalence of this condition. Further, if pressure ulcer severity is associated with additional increases in the length of stay, then a higher proportion of severe cases would likely be observed on a prevalence day. Then, if severe pressure ulcers are documented more often,7 8 the correlation between the administrative data and observed pressure ulcers would also appear to be higher. This cross-sectional sampling bias likely also explain why the total and acute length of stay for this sample are longer than the hospital's average length of stay (8.5 days, in 2012–2013).19

Our results should be considered in further iterations of the standards used to code pressure ulcers. In its current state, the ICD-10 mandates that diagnosis and diagnosis type are taken directly from the physician's documentation. Further investigation regarding how DAD coding and diagnosis typing for pressure ulcers can be more reflective of documentation by other healthcare professionals should be considered. Agreement on a common staging system for coding that aligns with the NPUAP guidelines would likely also improve the quality of the documentation. Further research is needed to understand if nursing documentation can improve the accuracy of pressure ulcer reporting and to design targeted quality improvement work.

Conclusion

The results of this study demonstrate that coding in the DAD may not accurately reflect the burden of hospital-acquired pressure ulcers. We recommend enhancements to the coding methods used to monitor and report on pressure ulcers by standardising the staging system used, and by considering the use of documented pressure ulcers by other healthcare professionals. With good benchmarking data, organisations will ultimately be able to design quality improvement strategies to prevent the development of hospital-acquired pressure ulcers and accurately evaluate the impact of the strategies.