The Physiological and Operative Severity Score for the enUmeration of Mortality and morbidity (POSSUM) (2024)

Search for other works by this author on:

Article history

Abstract

Introduction

Continuous personal professional audit is a requirement in modern surgical practice and patients increasingly want to know the results obtained by their surgeon. Monitoring crude death rates can mask the effects of case mix; surgeons who work in impoverished inner-city hospitals or tertiary referral centres may feel disadvantaged compared with their colleagues who elect to treat fit patients or work in affluent areas.

A number of scoring systems exist for the comparison of case mix1. The ideal risk scoring system should be able to predict morbidity and mortality, be quick, easy to use and should apply to all general surgical procedures. It should be applicable to any hospital and integrate easily into pre-existing audit programmes with minimal disruption2. The Physiological and Operative Severity Score for the enUmeration of Mortality and morbidity (POSSUM) is one such scoring system. The aim of this article is to describe the development of the POSSUM data set and the analytical methods it employs to compare the performance of individual surgeons. The benefits and disadvantages are discussed with respect to alternative comparative tools and with regard to the future role of POSSUM scoring in surgical practice.

Methods

A computerized search was conducted in three databases (Medline, the Cochrane Library and Embase) for the years 1990–2002. Relevant articles were then searched manually for further papers on risk analysis, case-mix comparison and POSSUM methodology. Although an extensive observational literature exists, no randomized trials were detected and the data must be interpreted with that in mind. The draft manuscript of this article was shown to a number of the clinicians who were instrumental in developing the POSSUM system (see Acknowledgements).

Development of the POSSUM data set

The POSSUM data set was developed over 2 years by Copeland et al.2. First, 35 possible risk factors for adverse outcome after operation were examined by multivariate analysis to detect which were predictive of death or complicated recovery. Only the 12 preoperative factors that were independently predictive of outcome were included in the final POSSUM data set. Each of the resulting 12 factors was graded and scored exponentially as 1, 2, 4 or 8, the highest being given to the most deranged value (Table 1). The POSSUM physiology score based on these preoperative factors was predictive of outcome for individual operations, but not for groups of surgical patients as a whole2. For example, a patient having an aortic aneurysm repair was likely to have a higher probability of death than the same patient having a pilonidal abscess drained. To address this, a six-factor operative severity score was added using similar methodology. Operative scores ranged from 1 to 8, depending on the size and severity of the operation (Table 2). Finally, the combined preoperative physiology and operative scores were subjected to logistic regression analysis to generate a risk equation that changed the scores into a predicted percentage mortality and morbidity. The system was tested prospectively for 6 months to confirm accuracy in predicting outcome after general surgical operations2.

Regression equations and methods of analysis

There is much confusion about the way POSSUM scores are analysed. The result of the POSSUM data set is a physiology score of 12–88 and an operative score of 6–44. Although the higher the overall POSSUM score, the greater the risk of morbidity and mortality, individual scores do not directly reflect the percentage risk. These two scores are most useful as part of a regression equation that is used to produce a percentage risk. The regression equation includes a constant number and weighted scores for physiology and operative risk (Fig. 1). A different constant and weighted value is needed to predict morbidity and mortality. This original system was used for the first 5 years, but many variations have since been published.

The initial equation used a new technique called exponential analysis; this works out the number of deaths in a certain risk group in the following way. The number of people in, for example, the band with a predicted risk of 50–60 per cent is calculated by working out the number with a risk greater than 50 per cent and subtracting the number with a risk of more than 60 per cent. This is not a standard statistical technique and has attracted criticism owing to the difficulty of giving a risk score to an individual. The number in a given range is also dependent on how many people are in other groups. Odd results occur occasionally, such as a negative predicted risk (negative number of deaths in a risk group). In this case the risk band is widened until it contains a positive number of deaths. Despite these criticisms, POSSUM scoring using the original equation and exponential analysis did predict outcome in general surgery2. However, later authors, particularly from Portsmouth, UK, could not reproduce the predictive power of the equation and criticized the analytical technique, although not the POSSUM data set.

Portsmouth POSSUM

An article by Whiteley et al.3 in 1996 showed how the original POSSUM regression equation failed to work in patients in Portsmouth. The authors found that POSSUM overpredicted death in a cohort of 1485 patients, particularly those at low risk. It was still possible to use the POSSUM data set, but a different regression equation was needed. This regression equation became known as the Portsmouth predictor equation, or P-POSSUM (Fig. 1). It used a different constant and weighted values for physiology and operative scores. P-POSSUM uses standard methods of analysis described by Hosmer and Lemeshow4. In this system, the risk applies to an individual. To work out the number of deaths in a given risk band, the average risk for each individual is calculated and multiplied by the number of people in the range. This is termed linear analysis.

A lively correspondence developed between the Portsmouth group and the originators of POSSUM in Manchester, UK. The initial letter5 questioned the statistical methods used by the Portsmouth group. Midwinter and Ashley6 also found that the Portsmouth regression equation more accurately predicted the outcome in their patients, most of whom had vascular procedures. The debate culminated in a direct comparison of the two possible methods of analysis. Wijesinghe et al.7 explained how the original POSSUM equation used exponential analysis, while P-POSSUM used linear analysis. They employed both methods in a series of 312 patients who had vascular surgical procedures and showed that each was effective if appropriate analysis was used7. It is important to understand that both systems fail when the incorrect analysis is used. Despite this article, subsequent authors have occasionally erred by using linear analysis with the original regression equation.

The Portsmouth group has continued to refine the regression equation, using the identical data set, and described a P-POSSUM equation that predicted outcome in 10 000 patient episodes in general surgery8. This was reported to be superior to the original POSSUM equation in their same data set, although the authors used linear, rather than exponential, analysis in their comparison. Copeland9 has recently published data showing that the original equation using exponential analysis has continued to be predictive in general surgery and, with some modifications, in orthopaedic surgery.

Pitfalls in data collection and analysis

Mistakes have been made in both data collection and analysis using POSSUM. Some authors have been criticized for recording their data in such a way as to alter the predicted mortality. Most of the POSSUM physiology and operative data set seems straightforward, but methods must be standard to be valid.

First, the POSSUM physiology score may change with time. For example, an elderly patient admitted with septicaemia from a diverticular abscess, who is aggressively resuscitated before operation, should have an improving physiology score. The authors of the original research used data taken as close to the time of operation as possible—the last recorded values before the patient entered the anaesthetic room. It could be argued that surgeons could improve their results artificially by selecting and recording the patient's worst physiology score. This does not matter as long as all surgeons agree to score at the same stage. There is room for further research into whether improving the physiology score by resuscitation improves overall outcome after operation. McIlroy et al.10 reported that preoperative resuscitation could improve physiology scores and that outcomes were poor in patients who failed to respond to resuscitation.

Missing data remain a problem11. In many cases, the results of some physiological data are not available. For some patients, tests such as chest radiography are not justified clinically. It is a misconception that such radiography is a required variable; raised jugulovenous pressure and shortness of breath at rest already give the highest cardiorespiratory score, and confirmation by imaging is not required. Investigations are performed on patients when there is reasonable suspicion that they may be abnormal. In the absence of a test result, the variable is currently scored at 1 (i.e. presumed normal).

There are also problems with specific data variables. The electrocardiogram (ECG) seems to cause the most confusion. A normal ECG scores 1, including normal variants such as right bundle branch block or sinus dysrhythmia. The middle category (score 2) includes slow atrial fibrillation or old myocardial damage. Recent myocardial infarction or evidence of myocardial ischaemia scores 8, the highest value. However, the highest score category also includes miscellaneous items. Confusion can occur easily if minor, non-specific ECG changes are scored in this miscellaneous category.

The operative score has an element of subjective assessment. The exact volume of blood loss may not be easy to determine, although the amount is scored in relatively broad bands. Peritoneal soiling has been inadvertently mis-scored in the past; some surgeons believe that all patients having elective repair of aortic aneurysm have maximum peritoneal contamination, but peritoneal soiling should be scored on opening the abdomen, and excludes blood. All of these problems could be eased by the creation of a comprehensive explanatory sheet for surgeons using POSSUM scoring.

Value of POSSUM in general surgery

The idea behind POSSUM was to enable a fair comparison between the performance of individual surgeons and individual hospitals. Most of the original POSSUM validation was carried out on general surgical procedures. The first paper2 modelled a general surgical population; it excluded both paediatric surgery, because physiological parameters in children differ from those of adults, and day-case surgery, where there is a low incidence of morbidity and death. POSSUM has since been used to compare the performance of individual surgeons12. In a study of 3006 general surgical episodes by five surgeons from a single hospital, crude mortality rates were compared with risk-adjusted outcomes (calculated from POSSUM observed : expected ratios). Individual surgeons had mortality rates that varied fivefold, from 1·0 to 4·9 per cent. Once adjusted, however, there was no significant difference in observed : expected mortality ratios12, which ranged from 0·86 to 1·06.

It has also been possible to compare the performance of individual hospitals13. For colorectal procedures, surgeons from a district general hospital were compared with those from a teaching hospital. In a 1-year study, mortality rates were 6 and 9 per cent in the two hospitals, but the morbidity rates were 9 and 26 per cent respectively, with the teaching hospital apparently performing significantly better. The POSSUM-predicted mortality rates for the two centres, however, were 5·2 and 9·8 per cent, whereas the predicted morbidity rates were 11·2 and 23·9 per cent respectively. It was concluded that both hospitals were performing as expected and that reporting of crude hospital mortality and morbidity rates could be misleading13.

POSSUM has also been used to compare surgical patients treated in different countries and appears to be valid in continents with healthcare systems different from that of the UK14–16. Even where resources, facilities and prehospital care differ, POSSUM can still predict outcome. Some of the sickest patients in hospital are managed in critical care facilities. In one study, the outcome of patients treated in a surgical high-dependency unit was predicted successfully by POSSUM scoring17. In this paper POSSUM outperformed the intensive care scoring system, Acute Physiology And Chronic Health Evaluation (APACHE) II, in predicting mortality. POSSUM scoring has also been used to assess the safety of transferring critically ill patients between the intensive care units of different hospitals16.

A criticism of POSSUM is that it applies only to surgical patients, and only to those who have an operation. Several authors have used the physiology component of the POSSUM score for patients who did not have a surgical procedure. In one study, 35 110 surgical admissions were analysed; the authors created a new regression equation that predicted mortality in the group, irrespective of whether an operation was performed or not18. The authors suggested that this could become a national minimum data set for all surgical admissions. It had the advantage of including patients too sick to undergo a surgical procedure.

The physiology component of the POSSUM score has been evaluated in some non-surgical procedures. In a study of patients who had intra-arterial thrombolysis for acute leg ischaemia, the POSSUM physiology score predicted mortality effectively19. Indeed, it is possible to predict mortality in surgical procedures, too, using only the physiology score; some of the best prediction equations in vascular surgical procedures were obtained without employing the operative data (V-POSSUM and V-POSSUM physiology only) (Fig. 1)20.

Value of POSSUM in specialist surgery

For most surgeons, their area of specialist expertise dictates their highest-risk operative procedures. Certain specialist surgeons were quick to see the advantages of POSSUM scoring as a way of allowing for case mix in their complex, high-risk operations. Cardiac surgeons were most involved following the well publicized report into perioperative paediatric cardiac mortality in Bristol, UK. Cardiac surgeons in the UK now collect data on all their procedures, and these are analysed using Bayesian methodology. An annual report is produced with risk-adjusted outcomes for individual cardiac units21. The UK government has recently decided that the risk-adjusted mortality rates for individual cardiac surgeons will be published and available in 2004. British cardiac surgeons have chosen not to use the POSSUM data set in their patients, but instead use a different data set of variables specific to their procedures.

Peripheral vascular surgeons are the next group to have felt exposed because of the number of major preventive operations (aortic aneurysm repair, carotid endarterectomy) they perform, for which mortality rates influence the justification for the procedure. Several vascular units have used POSSUM or P-POSSUM analysis to risk-stratify results for all arterial surgery22^,23, or for individual vascular procedures such as infrainguinal bypass and for elective24^,25 or emergency26^,27 aortic aneurysm repair.

The Vascular Surgical Society of Great Britain and Ireland (VSSGBI) has carried out several studies modelling outcome using the POSSUM data set with P-POSSUM analysis. It has also compared P-POSSUM and Bayesian analysis. The first study evaluated 1500 patients collected prospectively by 106 surgeons in 95 UK hospitals20^,22. The investigation showed that P-POSSUM accurately predicted outcome in a variety of arterial procedures across a range of hospitals and surgeons. In addition, it was possible to create useful models using the physiological data alone20. Subsequent articles have challenged the use of a generic equation to model patients having carotid endarterectomy (very low mortality rate)28 and ruptured abdominal aortic aneurysm (RAAA) surgery (very high mortality rate), where unique regression equations are preferable (RAAA-POSSUM and RAAA-POSSUM physiology only) (Fig. 1)26^,27. It is an inescapable conclusion that separate equations will probably be required for each major procedure29. The VSSGBI currently collects data nationally on three index procedures (carotid endarterectomy, aortic aneurysm repair and infrainguinal bypass) using the POSSUM data set, and is building up sufficient data to generate risk models for each procedure.

Coloproctologists in the UK have only recently begun to collect data for risk stratification. The Association of Coloproctologists of Great Britain and Ireland has provided a minimum data set and many centres have added extra data that they believe to be important. The data set includes POSSUM data items but it will also be analysed using Bayesian methods. An initial study has been carried out to compare outcomes after emergency colorectal surgery. The Association of Upper Gastrointestinal Surgeons in the UK intends to form a similar registry to those of the cardiac and vascular surgeons for all patients undergoing operation for oesophageal, stomach and pancreatic cancer. It is expected that POSSUM and Bayesian analysis will be used to stratify case mix.

Where POSSUM should be used with caution

POSSUM has been modelled on 30-day mortality. Care must be exercised whenever an outcome other than 30-day mortality is examined with POSSUM; in-hospital mortality rates may be significantly different from 30-day rates30. Care must also be used when POSSUM is employed to analyse new procedures. The use of endovascular aortic aneurysm repair was justified by a lower predicted risk than that for open repair25. POSSUM scoring may predict 30-day outcome, but this does nothing to inform us about the mid- to long-term success of a procedure.

POSSUM has also been used retrospectively to justify mortality in a diagnostic procedure. Wang and Tu31 used the physiological POSSUM score as one measure of patient sickness to examine four deaths after barium enema procedures. Although barium enema is associated with occasional death, a huge number of procedures would be required to show that POSSUM could predict it.

Comparison of POSSUM with other scoring systems

APACHE is the scoring system used most extensively in critical care. The original system created in 1981 had two parts, both of which measured physiological variables: the acute physiology score has 34 variables that measure the acute disturbance, while the chronic health evaluation score reflects a patient's level of function before the illness. The variables were chosen by a panel of clinicians based on their perceived importance, rather than logistic regression32. The original APACHE score was effective in risk stratification but was unwieldy to use with so many variables to collect. In 1985, APACHE II was developed to simplify the system; 34 variables were reduced to 12. APACHE II was never intended to predict death in an individual patient, but some authors have used it for this purpose. In critical care, one use of APACHE II involved serial measurement to monitor changes in response to treatment, but this is still associated with a tendency to overestimate the risk of death33^,34. APACHE II has been validated in a much wider range of patients than POSSUM, including medical and surgical patients admitted to intensive care, and those who did not undergo operation.

The only direct comparison between the two systems included 117 patients; it showed that POSSUM was more predictive of outcome than APACHE II17. The POSSUM data set is smaller and was created by logistic regression rather than by suggestions from a panel of experts32. APACHE III, introduced in 199135, attempted to correct some of these problems by using statistical techniques to choose a data set. There is no published work comparing its power with that of POSSUM.

Other outcome prediction systems that are easier to use because they have relatively few factors have not been discriminating enough. Individual factors such as intra-abdominal pressure36, platelet count37, sigmoid mucosal pH38 and even the patient's signature39 have been used with limited success. Several authors have tried to produce simpler classification systems. The Hardman index consists of five clinical indicators to predict death after repair of a ruptured aortic aneurysm. In a study that compared Hardman and POSSUM scoring in 191 patients with a ruptured aortic aneurysm, POSSUM predicted the risk of death after operation in broader bands that were more clinically useful40. Another recent system is the Surgical Risk Scale, which has the advantage of being discriminating for low-risk procedures, but which requires further validation for high-risk procedures41.

Both cardiac and vascular surgeons have explored the use of Bayesian analysis as a means of comparing risk-adjusted outcome. Bayes is a learning database that uses experience of data collected to predict the probability of an outcome for an individual with particular risk characteristics. It then monitors its own accuracy and updates its analysis as more data accrue. Surgeons initially choose variables likely to predict outcome; the Bayes database then ‘learns’ which variables are important and gives more weight to them. It is then possible to determine predicted mortality and to compare results of an individual surgeon with those from the database provided by his or her peers. The VSSGBI has compared Bayes and P-POSSUM for a large number of various vascular procedures22 and for three index vascular procedures42. Both methods were effective, and little difference was found, although the database remains immature42.

Discussion

It is now mandatory for surgeons to audit their clinical work. Simple counting of postoperative deaths and complications is not fair and produces misleading comparisons between surgeons, clinical teams and hospitals. An adjustment for case mix is needed and POSSUM scoring is one method that has been explored over the past decade. Most clinicians are familiar with the data items collected in the physiological and operative data sets, but there seems to be a bewildering array of regression equations and techniques of analysis. In fact, the choice is relatively limited. To monitor general surgical procedures between surgeons and hospitals, the original POSSUM equation using exponential analysis should be used2^,7^,43. If the standard linear methodology is used for analysis, then a P-POSSUM equation must be employed7.

Vascular surgeons have validated P-POSSUM methodology, although a different regression equation is likely to emerge for each of the index procedures. POSSUM methodology, and in particular the data set, has been used successfully to compare the performance of general and specialist surgeons between hospitals and across continents. Surgeons who publish comparative data must, however, document explicitly the regression equation and statistical method of analysis used to obtain observed : expected outcome results.

POSSUM has acknowledged limitations; the equations perform poorly in predicting death in procedures that have a very low associated mortality rate. Low-risk, high-volume operations are often better audited using process measures rather than outcome. Similarly, it predicts only 30-day mortality. The use of POSSUM to predict long-term outcome has yet to be validated. Still, of all the risk-scoring systems currently available, there is most experience with POSSUM in both specialist and non-specialist surgery. Items in the POSSUM data set are commonly collected and are now in many routine hospital databases. Other systems that require invasive tests (e.g. blood gases) must be used with caution if these investigations are not part of routine care; no procedure is without complication.

Finally, the results of comparative audit with POSSUM cannot be used to attribute blame. Patient care is the responsibility of individual surgeons, but outcomes often depend on a large multidisciplinary team comprising surgeons, anaesthetists, intensive care staff, junior doctors and nurses, all of whom may affect complication and death rates29. It is unhelpful to look at complications purely in terms of surgical blame (anastomotic leaks can be due to poor nutrition) or anaesthetic blame (myocardial infarction may be precipitated by the stress of a prolonged operation). Where audit shows a change in mortality rates and a significant increase in the observed : expected ratio, the practice of the entire team should be reviewed. POSSUM is simply a tool for fair comparative audit and its methodology currently stands comparison with other sophisticated methods of case-mix analysis, such as APACHE and Bayes.

Acknowledgements

The authors thank Mr Graham Copeland and Dr David Prytherch for helpful advice in the preparation of this manuscript.

References

Copyright © 2003 British Journal of Surgery Society Ltd. Published by John Wiley & Sons, Ltd.

Issue Section:

Review

Email alerts

See also

Citing articles via

More from Oxford Academic