Comparison of the National Early Warning Score in non-elective medical and surgical patients.

BACKGROUND: The National Early Warning Score (NEWS) is used to identify deteriorating patients in hospital. NEWS is a better discriminator of outcomes than other early warning scores in acute medical admissions, but it has not been evaluated in a surgical population. The study aims were to evaluate the ability of NEWS to discriminate cardiac arrest, death and unanticipated ICU admission in patients admitted to surgical specialties, and to compare the performance of NEWS in admissions to medical and surgical specialties. METHODS: Hospitalwide data over 31 months, from adult inpatients who stayed at least one night or died on the day of admission, were analysed. The data were categorized as elective or non-elective surgical or medical admissions. The ability of NEWS to discriminate the outcomes above in these different groups was assessed using the area under the receiver operating characteristic curve (AUROC). RESULTS: There were too few outcomes to permit meaningful comparison of elective admissions, so the analysis was constrained to comparison of non-elective admissions. NEWS performed equally well, or better, for surgical as for medical patients. For death within 24 h the AUROC for surgical admissions was 0·914 (95 per cent c.i. 0·907 to 0·922), compared with 0·902 (0·898 to 0·905) for medical admissions. For the combined outcome of any of death, cardiac arrest or unanticipated ICU admission, the AUROC was 0·874 (0·868 to 0·880) for surgical admissions and 0·874 (0·871 to 0·877) for medical admissions. CONCLUSION: NEWS discriminated deterioration in non-elective surgical patients at least as well as in non-elective medical patients.

Predicting outcome following colorectal cancer surgery using a colorectal biochemical and haematological outcome model (Colorectal BHOM).

AIM: To present a new biochemistry and haematology outcome model which uses a minimum dataset to model outcome following colorectal cancer surgery, a concept previously shown to be feasible with arterial operations. METHOD: Predictive binary logistic regression models (a mortality and morbidity model) were developed for 704 patients who underwent colorectal cancer surgery over a 6-year period in one hospital. The variables measured included 30-day mortality and morbidity. Hosmer-Lemeshow goodness of fit statistics and frequency tables compared the predicted vs the reported number of deaths. Discrimination was quantified using the c-index. RESULTS: There were 573 elective and 131 nonelective interventional cases. The overall mean predicted risk of death was 7.79% (50 patients). The actual number of reported deaths was also 50 patients (χ(2) = 1.331, df = 4, P-value = 0.856; no evidence of lack of fit). For the mortality model, the predictive c-index was = 0.810. The morbidity model had less discriminative power but there was no evidence of lack of fit (χ(2) = 4.198, df = 4, P-value = 0.380, c-index = 0.697). CONCLUSIONS: The Colorectal Biochemistry and Haematology Outcome mortality model suggests good discrimination (c-index > 0.8) and uses only a minimal number of variables. However, it needs to be tested on independent datasets in different geographical locations.

The development of a VBHOM-based outcome model for lower limb amputation performed for critical ischaemia.

BACKGROUND: VBHOM (Vascular Biochemistry and Haematology Outcome Models) adopts the approach of using a minimum data set to model outcome and has been previously shown to be feasible after index arterial operations. This study attempts to model mortality following lower limb amputation for critical limb ischaemia using the VBHOM concept. METHODS: A binary logistic regression model of risk of mortality was built using National Vascular Database items that contained the complete data required by the model from 269 admissions for lower limb amputation. The subset of NVD data items used were urea, creatinine, sodium, potassium, haemoglobin, white cell count, age on and mode of admission. This model was applied prospectively to a test set of data (n=269), which were not part of the original training set to develop the predictor equation. RESULTS: Outcome following lower limb amputation could be described accurately using the same model. The overall mean predicted risk of mortality was 32%, predicting 86 deaths. Actual number of deaths was 86 (chi(2)=8.05, 8 d.f., p=0.429; no evidence of lack of fit). The model demonstrated adequate discrimination (c-index=0.704). CONCLUSIONS: VBHOM provides a single unified model that allows good prediction of surgical mortality in this high risk group of individuals. It uses a small, simple and objective clinical data set that may also simplify comparative audit within vascular surgery.

POSSUM models in open abdominal aortic aneurysm surgery.

OBJECTIVES: This study evaluated the Physiological and Operative Severity Score for the enUmeration of Mortality and Morbidity (POSSUM), Portsmouth (P) POSSUM and Vascular (V) POSSUM. The primary aim was to assess the validity of these scoring systems in a population of patients undergoing elective and emergency open AAA repair. The secondary intention was in the event that these equations did not fit all patients with an aneurysm; a new model would be developed and tested using logistic regression from the local data (Cambridge POSSUM). METHODS: POSSUM data items were collected prospectively in a group of 452 patients undergoing elective and emergency open AAA repair over an eight-year period. The operative mortality rates were compared with those predicted by POSSUM, P-POSSUM, V-POSSUM and Cambridge POSSUM. RESULTS: All models except V-POSSUM (physiology only) showed significant lack of fit when predicting mortality after open AAA surgery. It was found that the locally generated single unified model (Cambridge POSSUM) could successfully describe both elective and ruptured AAA mortality with good discrimination (chi(2)=9.24, 7 d.f., p=0.236, c-index=0.880). CONCLUSIONS: POSSUM, V-POSSUM and P-POSSUM may not be robust tools for comparing mortality between populations undergoing elective and emergency open AAA repair as once thought. The development and successful validation of Cambridge POSSUM provides a unified model to describe both elective and emergency AAAs together and should be validated in other geographical settings.

VBHOM, a data economic model for predicting the outcome after open abdominal aortic aneurysm surgery.

BACKGROUND: Vascular Biochemistry and Haematology Outcome Models (VBHOM) adopted the approach of using a minimum data set to model outcome. This study aimed to test such a model on a cohort of patients undergoing open elective and non-elective abdominal aortic aneurysm (AAA) repair. METHODS: A binary logistic regression model of risk of in-hospital mortality was built from the 2002-2004 submission to the UK National Vascular Database (NVD) (2718 patients). The subset of NVD data items used comprised serum levels of urea, sodium and potassium, haemoglobin, white cell count, sex, age and mode of admission. The model was applied prospectively using Hosmer-Lemeshow methodology to a test data set from the Cambridge Vascular Unit. RESULTS: The validation set contained 327 patients, of whom 208 had elective AAA repair and 119 had emergency repair of a ruptured AAA. Outcome following elective and non-elective AAA repair could be described accurately using the same model. The overall mean predicted risk of death was 14.13 per cent, and 48 deaths were predicted. The actual number of deaths was 53 (chi(2) = 8.40, 10 d.f., P = 0.590; no evidence of lack of fit). The model also demonstrated good discrimination (c-index = 0.852). CONCLUSION: The VBHOM approach has the advantage of using simple, objective clinical data that are easy to collect routinely. The VBHOM data items potentially allow prediction of risk in an individual patient before aneurysm surgery.

Measuring clinical performance using routinely collected clinical data.

Following the well-publicized problems with paediatric cardiac surgery at the Bristol Royal Infirmary, there is wide public interest in measures of hospital performance. The Kennedy report on the BRI events suggested that such measures should be meaningful to the public, case-mix-adjusted, and based on data collected as part of routine clinical care. We have found that it is possible to predict in-hospital mortality (a measure readily understood by the public) using simple routine data-age, mode of admission, sex, and routine blood test results. The clinical data items can be obtained at a single venesection, are commonly collected in the routine care of patients, are already stored on hospital core IT systems, and so place no extra burden on the clinical staff providing care. Such risk models could provide a metric for use in evidence-based clinical performance management. National application is logistically feasible.

The use of routine laboratory data to predict in-hospital death in medical admissions.

The ability to predict clinical outcomes in the early phase of a patient's hospital admission could facilitate the optimal use of resources, might allow focused surveillance of high-risk patients and might permit early therapy. We investigated the hypothesis that the risk of in-hospital death of general medical patients can be modelled using a small number of commonly used laboratory and administrative items available within the first few hours of hospital admission. Matched administrative and laboratory data from 9497 adult hospital discharges, with a hospital discharge specialty of general medicine, were divided into two subsets. The dataset was split into a single development set, Q(1) (n=2257), and three validation sets, Q(2), Q(3) and Q(4) (n(1)=2335, n(2)=2361, n(3)=2544). Hospital outcome (survival/non-survival) was obtained for all discharges. An outcome model was constructed from binary logistic regression of the development set data. The goodness-of-fit of the model for the validation sets was tested using receiver-operating characteristics curves (c-index) and Hosmer-Lemeshow statistics. Application of the model to the validation sets produced c-indices of 0.779 (Q(2)), 0.764 (Q(3)) and 0.757 (Q(4)), respectively, indicating good discrimination. Hosmer-Lemeshow analysis gave chi(2)=9.43 (Q(2)), chi(2)=7.39 (Q(3)) and chi(2)=8.00 (Q(4)) (p-values of 0.307, 0.495 and 0.433) for 8 degrees of freedom, indicating good calibration. The finding that the risk of hospital death can be predicted with routinely available data very early on after hospital admission has several potential uses. It raises the possibility that the surveillance and treatment of patients might be categorised by risk assessment means. Such a system might also be used to assess clinical performance, to evaluate the benefits of introducing acute care interventions or to investigate differences between acute care systems.

Risk-adjusted predictive models of mortality after index arterial operations using a minimal data set.

BACKGROUND: Reducing the data required for a national vascular database (NVD) without compromising the statistical basis of comparative audit is an important goal. This work attempted to model outcomes (mortality and morbidity) from a small and simple subset of the NVD data items, specifically urea, sodium, potassium, haemoglobin, white cell count, age and mode of admission. METHODS: Logistic regression models of risk of adverse outcome were built from the 2001 submission to the NVD using all records that contained the complete data required by the models. These models were applied prospectively against the equivalent data from the 2002 submission to the NVD. RESULTS: As had previously been found using the P-POSSUM (Portsmouth POSSUM) approach, although elective abdominal aortic aneurysm (AAA) repair and infrainguinal bypass (IIB) operations could be described by the same model, separate models were required for carotid endarterectomy (CEA) and emergency AAA repair. For CEA there were insufficient adverse events recorded to allow prospective testing of the models. The overall mean predicted risk of death in 530 patients undergoing elective AAA repair or IIB operations was 5.6 per cent, predicting 30 deaths. There were 28 reported deaths (chi(2) = 2.75, 4 d.f., P = 0.600; no evidence of lack of fit). Similarly, accurate predictions were obtained across a range of predicted risks as well as for patients undergoing repair of ruptured AAA and for morbidity. CONCLUSION: A 'data economic' model for risk stratification of national data is feasible. The ability to use a minimal data set may facilitate the process of comparative audit within the NVD.

Development of a dedicated risk-adjustment scoring system for colorectal surgery (colorectal POSSUM).

BACKGROUND: The aim of the study was to develop a dedicated colorectal Physiological and Operative Severity Score for the enUmeration of Mortality and morbidity (CR-POSSUM) equation for predicting operative mortality, and to compare its performance with the Portsmouth (P)-POSSUM model. METHODS: Data were collected prospectively from 6883 patients undergoing colorectal surgery in 15 UK hospitals between 1993 and 2001. After excluding missing data and 93 patients who did not satisfy the inclusion criteria, 4632 patients (68.2 per cent) underwent elective surgery and 2107 had an emergency operation (31.0 per cent); 2437 operations (35.9 per cent) for malignant and 4267 (62.8 per cent) for non-malignant diseases were scored. Stepwise logistic regression analysis was used to develop an age-adjusted POSSUM model and a dedicated CR-POSSUM model. A 60:40 per cent split-sample validation technique was adopted for model development and testing. Observed and expected mortality rates were compared. RESULTS: The operative mortality rate for the series was 5.7 per cent (387 of 6790 patients) (elective operations 2.8 per cent; emergency surgery 12.0 per cent). The CR-POSSUM, age-adjusted POSSUM and P-POSSUM models had similar areas under the receiver-operator characteristic curves. Model calibration was similar for CR-POSSUM and age-adjusted POSSUM models, and superior to that for the P-POSSUM model. The CR-POSSUM model offered the best overall accuracy, with an observed : expected ratio of 1.000, 0.998 and 0.911 respectively (test population). CONCLUSION: The CR-POSSUM model provided an accurate predictor of operative mortality. External validation is required in hospitals different from those in which the model was developed.

Risk-adjusted prediction of operative mortality in oesophagogastric surgery with O-POSSUM.

INTRODUCTION: The present study was designed to develop a dedicated oesophagogastric model for the prediction of risk-adjusted postoperative mortality in upper gastrointestinal surgery (O-POSSUM). METHODS: Using 1042 patients undergoing oesophageal (n = 538) or gastric (n = 504) surgery between 1994 and 2000 the Portsmouth predictor equation for mortality (P-POSSUM) scoring system was compared with a standard logistic regression O-POSSUM model and a multilevel O-POSSUM model using the following independent factors: age, physiological status, mode of surgery, type of surgery and histological stage. RESULTS: The overall mortality rate was 12.0 per cent (elective mortality rate 9.4 per cent and emergency mortality rate 26.9 per cent). P-POSSUM overpredicted mortality (14.5 per cent), particularly in the elective group of patients. The multilevel model offered higher discrimination than the single-level O-POSSUM and P-POSSUM models (area under receiver-operator characteristic curve 79.7 versus 74.6 and 74.3 per cent). When observed to expected outcomes were evaluated, the multilevel O-POSSUM model was found to offer better calibration (Hosmer-Lemeshow chi(2) statistic 10.15 versus 10.52 and 28.80). CONCLUSION: The multilevel O-POSSUM model provided an accurate risk-adjusted prediction of death from oesophageal and gastric surgery for individual patients. In conjunction with a multidisciplinary approach to patient management, the model may be used in everyday practice for perioperative counselling of patients and their carers.

Comparison of P-POSSUM risk-adjusted mortality rates after surgery between patients in the USA and the UK.

BACKGROUND: The Physiological and Operative Severity Score for the enUmeration of Mortality and morbidity (POSSUM) criteria have been used to assess surgical risk in patients in the UK. The aim was to determine how applicable these criteria are to patients undergoing surgery in the USA. METHODS: Two cohorts of patients undergoing major non-cardiac surgery were followed prospectively in the USA (n = 1056) and the UK (n = 1539). Each patient was assigned a risk score for preoperative physiological status and operative severity using the established POSSUM criteria. Death in hospital was the primary outcome measure. For each patient a predicted risk of death was calculated from Portsmouth POSSUM (P-POSSUM) methodology using an established equation. The relationships between predicted and observed mortality rates in each cohort were investigated by means of multivariate logistic regression. RESULTS: Within each cohort, an increase in risk estimated by P-POSSUM predicted an increase in observed mortality rate (P < 0.001). For any given risk level, however, mortality rates were significantly higher in the UK cohort than in the US cohort (odds ratio 4.50 (95 per cent confidence interval 2.81 to 7.19); Z = 6.25, P < 0.001). CONCLUSION: An increase in predicted risk, based on the P-POSSUM methodology, was associated with a higher mortality rate in patients from both countries. However, risk-adjusted mortality rates following major surgery were four times higher in the UK cohort. These marked differences warrant validation in a larger number of centres.

Towards a national clinical minimum data set for general surgery.

BACKGROUND: Measurement and comparison of surgical performance is accepted as necessary and inevitable. Risk-stratified (case-mix adjusted) models of clinical outcomes form a metric with which to assess performance, but require accurate data. Collecting such data in the clinical environment is time consuming and difficult. This study aimed to construct effective models, for operative and non-operative admissions, from routine clinical data residing in hospital computers, so minimizing data collection and quality problems, and facilitating national implementation. METHODS: Data for 3181 non-operative emergency, 5039 elective and 3043 emergency operative admissions for the 2 years beginning 1 August 1997 were used to generate logistic regression equations for risk of death, which were applied prospectively to the following 3 years' data. RESULTS: The models use urea, haemoglobin, white blood cell count, sodium, potassium, age on admission, sex, British United Provident Association (BUPA) Operative Severity Score (for operative admissions) and, implicitly, mode of admission and mortality at discharge. All three models successfully stratified risk into five or more bands. CONCLUSION: Effective models of mortality, applicable to all general surgical admissions, can be constructed from existing routine clinical data, largely obtained from a single venesection. The data set is a candidate national clinical minimum data set.

Portsmouth POSSUM models for abdominal aortic aneurysm surgery.

BACKGROUND: The Portsmouth Physiological and Operative Severity Score for enUmeration of Mortality and morbidity (P-POSSUM) equation permits comparative audit that allows for differences in case mix. The methodology has previously been shown accurately to model general surgical and vascular surgical patients. Patients with a ruptured abdominal aortic aneurysm (AAA) are, however, very different from elective patients and it may be hypothesized that they require their own specific risk model. METHODS: POSSUM data on 444 (213 emergency, 231 elective) admissions for AAA surgery between August 1993 and July 2000 were analysed using the P-POSSUM equation for general surgery and the P-POSSUM equations developed for vascular surgery. RESULTS: All equations successfully modelled the elective aneurysms but failed to fit the emergency aneurysms, and the elective and emergency aneurysms combined. This suggested that admission method (not a POSSUM data item) is an important factor for patients with AAA. However, with these data it was not possible to generate a model, including admission method, that successfully modelled the combined elective and emergency data. The 213 emergency aneurysm repairs were divided into two groups by operation date. The first 106 (training set) were used to form logistic regression models following the P-POSSUM methodology. These models were found successfully to fit the remaining 107 records (test set) on prospective application (chi2 = 4.50, 4 d.f., P = 0.345). CONCLUSION: Ruptured AAAs appear to be different from elective AAAs and other vascular cases and require their own risk model.

A model for national outcome audit in vascular surgery.

OBJECTIVE: The aim was to model vascular surgical outcome in a national study using POSSUM scoring. METHODS: One hundred and twenty-one British and Irish surgeons completed data questionnaires on patients undergoing arterial surgery under their care (mean 12 patients, range 1-49) in May/June 1998. A total of 1480 completed data records were available for logistic regression analysis using P-POSSUM methodology. Information collected included all POSSUM data items plus other factors thought to have a significant bearing on patient outcome: "extra items". The main outcome measures were death and major postoperative complications. The data were checked and inconsistent records were excluded. The remaining 1313 were divided into two sets for analysis. The first "training" set was used to obtain logistic regression models that were applied prospectively to the second "test" dataset. RESULTS: using POSSUM data items alone, it was possible to predict both mortality and morbidity after vascular reconstruction using P-POSSUM analysis. The addition of the "extra items" found significant in regression analysis did not significantly improve the accuracy of prediction. It was possible to predict both mortality and morbidity derived from the preoperative physiology components of the POSSUM data items alone. CONCLUSION: this study has shown that P-POSSUM methodology can be used to predict outcome after arterial surgery across a range of surgeons in different hospitals and could form the basis of a national outcome audit. It was also possible to obtain accurate models for both mortality and major morbidity from the POSSUM physiology scores alone.

POSSUM and Portsmouth POSSUM for predicting mortality. Physiological and Operative Severity Score for the enUmeration of Mortality and morbidity.

BACKGROUND: There is a need for an accurate measure of surgical outcomes so that hospitals and surgeons can be compared properly regardless of case mix. POSSUM (Physiological and Operative Severity Score for the enUmeration of Mortality and morbidity) uses a physiological score and an operative severity score to calculate risks of mortality and morbidity. In a previous small study it was found that Portsmouth POSSUM (P-POSSUM; a modification of the POSSUM system) provided a more accurate prediction of mortality. METHODS: Some 10000 general surgical interventions (excluding paediatric and day cases) were studied prospectively between August 1993 and November 1995. The POSSUM mortality equation was applied to the full 10000 surgical episodes. The 10000 patients were arranged in chronological order and the first 2500 were used as a training set to produce the modified P-POSSUM predictor equation. This was then applied prospectively to the remaining 7500 patients arranged chronologically in five groups of 1500. RESULTS: The original POSSUM logistic regression equation for mortality overpredicts the overall risk of death by more than twofold and the risk of death for patients at lowest risk (5 per cent or less) by more than sevenfold. The P-POSSUM equation produced a very close fit with the observed in-hospital mortality. CONCLUSION: P-POSSUM provides an accurate method for comparative surgical audit.

An evaluation of the POSSUM surgical scoring system.

POSSUM (Physiological and Operative Severity Score for the enUmeration of Morbidity and mortality) has been studied as a possible surgical audit system for a 9-month interval using a sample of 28 per cent of the general surgical workload. Mortality or survival was analysed as an endpoint. In this sample the published POSSUM predictor equation for mortality overpredicted deaths by a factor of more than two. The bulk of the overprediction occurred in the group at lowest risk (predicted mortality 10 per cent or less), in which death was overpredicted by a factor of six. This is the most important group for audit purposes since it contains the majority of surgical patients and is composed of fit patients undergoing minor surgery. The published predictor equation for mortality returns a minimum predicted mortality of 1.08 per cent, clearly far higher than that expected for a fit patient having minor surgery. Logistic regression was done on a set of 1485 surgical episodes to generate a local predictor equation for mortality. This process gave a predictor equation that fitted well with the observed mortality rate and gave a minimum predicted risk of mortality of 0.20 per cent. The previously published POSSUM predictor equation for mortality performed badly when tested using a standard test of goodness of fit for logistic regression and must be modified.