Introduction

The global population is evolving with a progressively increasing life expectancy. The geriatric population is the fastest growing subgroup in most developed countries. In 2017, the average life expectancy in Sweden was 84 and 81 years for women and men, respectively [1]. The aging population challenges healthcare services. Increasing age increases the risk of post-operative complications and mortality following surgery [2]. The risk of death after surgery has been shown to increase in the presence of a post-operative complication, which is more significant than a high pre-operative risk profile [3,4]. In addition, after traumatic insults, geriatric patients have a higher mortality risk for a longer period of time [5,6]. Patients over the age of seventy have demonstrated a mortality rate of over 20% following emergency laparotomy [7]. The causality profile for such observations is complex but is likely due to factors including comorbidity burden, weakened physiological reserves and general frailty.

In contemporary healthcare systems, geriatric patients frequently undergo emergency general surgery. It is therefore highly relevant to validate robust outcome predictors of this patient group in order to avoid futile care measures. In this vulnerable group of patients, abdominal surgery may result in complications and extended intensive care stays. It is therefore vital to be able to provide pre-operative predictors of survival in order to facilitate valid medical decisions. Historically, this process has used age as a marker of general physical strength and thereby likelihood of survival. Nevertheless, age suffers many limitations as a major predictor of outcome following surgery [2]. In recent years, several scoring instruments have been developed to predict mortality of the surgical patient including POSSUM (Physiological and Operative Severity Score for the enumeration of Mortality and morbidity) [8] which was modified, improved and renamed P-POSSUM (Portsmouth - POSSUM) [9], ASA (American Society of Anaesthesiology) [10], APACHE (Acute Physiology and Chronic Health Evaluation) [11], and physical frailty as measured through osteopenia and sarcopenia [12,13]. Previous studies have demonstrated both strengths and weaknesses in these instruments, however, as of yet there is no consensus as to which one is more reliable in the geriatric population or how the instruments compare to each other in mortality predictive ability. This study aims to assess the performance of P-POSSUM and predictors of frailty in calculating 30-day mortality for geriatric patients undergoing emergency laparotomy. The study hypothesis was a significant correlation between P-POSSUM and osteopenia with poor outcomes.

Materials and Methods

Patient Inclusion

Ethical approval for this study was obtained from the institutional review board of Uppsala County (Ref. 2017/421). All geriatric patients (65 years of age and older) undergoing emergency laparotomy were identified between January 1st, 2015 and December 31st, 2016 from the surgical registry of Orebro University Hospital in Orebro, Sweden. Hospital electronic medical records were used to obtain information about age, sex, body mass index (BMI), diagnosis according to the International Statistical Classification of Diseases (ICD) 10th version, blood tests and vital signs on admission, transfusion requirements, Charlson Comorbidity Index (CCI) score, American Society of Anaesthesiologists (ASA) classification, post-operative complications, hospital length of stay (LOS), and 30-day mortality. Primary outcome of interest was to assess the predictive ability of P-POSSUM, osteopenia, and sarcopenia.

Radiology

Osteopenia and sarcopenia were assessed by a consultant radiologist using the most recent (but never older than 30 days prior to the date of surgery) computed tomography (CT) for measurements of bone density and muscle area. If both a low-dose and a normal-dose radiation CT were performed within 30 days before surgery, the normal-dose CT was chosen. Most studies were performed with intravenous contrast media. Sarcopenia was measured as total skeletal muscle area in a transaxial CT slice, 3 or 5 mm thick, at the L3 vertebral level. The computer programme ImageJ was used for segmentation of all tissue with an attenuation between -29 and 150 Hounsfield Units (HU). When the segmentation erroneously included other tissue such as bowel or subcutaneous oedema, this was edited manually with the CT image as reference. For sarcopenia, skeletal muscle index thresholds of 52.4 cm²/m² for men and 38.5 cm²/m² for women were used [14]. Body surface area was calculated from weight and height according to the Du Bois formula [15]; body surface area (m²) = 0.007184 × W^0.425 × H^0.725. In the same CT slice, the bone density was measured as the average HU in a 1.5-2 cm² region of interest in the trabecular part of the L3 vertebra. Osteopenia was defined as HU < 100¹².

Statistical Methods

Patients demographics and clinical outcomes were analysed using descriptive statistical methods. Continuous data are summarized as mean (standard deviation, SD) or median [interquartile range, IQR], wherever suitable, and categorical and ordinal data are presented as percentages. Differences between groups were tested using Student’s t-test or Mann-Whitney U test for continuous or ordinal data, wherever suitable, and using c² test or Fisher’s exact test for categorical data. A log transformation was carried out for the following variables: C-reactive protein (CRP), creatinine, and P-POSSUM for mortality and morbidity.

Associations between 30-day mortality and patient covariates were evaluated using Poisson regression model with robust standard errors. Potential confounding was adjusted for by including the following covariates in the model: gender, age, known cardiac or pulmonary conditions, the presence of a cancer diagnosis, Charlson comorbidity index (CCI), BMI, heart rate, systolic blood pressure, haemoglobin levels, log transformed CRP (logCRP) and creatinine (logCreatinine), number of procedures, requirement for packed red blood cell (PRB) transfusion, indication for surgery and surgical procedure, physiology score, operative severity score, log transformed morbidity and mortality P-POSSUM (logMorbidity P-POSSUM and logMortality P-POSSUM), osteopenia, ASA classification, surgery indication and procedure. Only a handful of patients from the complete data set met the criteria for sarcopenia. Therefore, this variable and its potential role as a possible predictor of mortality was excluded from the analyses. Variable selection was also conducted using a backward stepwise method with significance level of 0.10 for removal from the regression model and significance level of 0.05 for addition to the regression model. For categorical or ordinal variables presented as dummy variables in the stepwise regression analysis, the original variable would be kept in the final model if any of the derived dummy variables were found to be statistically significant.

In Poisson regression analysis, missing data were multiply imputed by using the iterative Markov chain Monte Carlo method. Five imputed data sets were generated. The estimates from the five data sets were combined according to Rubin’s rules [16]. A two-sided p value of less than 0.05 was considered statistically significant. The prognostic value of P-POSSUM and osteopenia were evaluated using sensitivity, specificity and area under receiver operating characteristic (ROC) curve. Youden’s J statistic was used to find optimum cut-off point on ROC curve. Difference between the areas under ROC curves (AUCs) was tested using c² test. All analyses were conducted in the statistical software packages R 3.5.0 (R Core Team, R Foundation for Statistical Computing, Vienna, Austria) and Stata 15.1 (StataCorp, College Station, TX, USA).

Results

In total 209 patients with a mean age of 75.6 (7.3) years were included in the study. There was an equal split between men and women. The average CCI score was 5.77 (2.36) and the most common ASA class was 3 (53.3%). The most common indication for surgery was bowel obstruction which occurred in 119 (56.9%) patients and the most common surgical procedure during laparotomy was bowel resection occurring in 82 patients (39.2%). Post-operative complications ranged between 5.3%-36.1% with infection being the most common type of complication (Table 1). Median [IQR] length of hospital stay was 12 (7, 21) days and the 30-day mortality rate was 31.1% (Table 1). When comparing patients who died within 30 days of surgery to those who survived beyond this point, there were some significant differences. Patients who died within 30 days were on average slightly older (78.66 vs. 74.19 years, p <0.001), had a higher average CCI (6.63 vs. 5.38, p <0.001), higher frequency of ASA class ≥ 4 (30.4% vs. 10.1%, p <0.001), higher average creatinine levels on admission (108.50 vs. 78.00 µmol/L, p = 0.003), higher PRB transfusion requirements (58.5% vs. 32.6%, p = 0.001) and higher proportion of osteopenia (59.0% vs. 41.0%, p = 0.029).

Table 1. Demographics and clinical outcome split into total cohort, patients who survived beyond 30 days of surgery and patients who died <30 days of surgery

^aASA: American Society of Anaesthesiologists; ^bBMI: Body Mass Index; ^cGCS: Glasgow Coma Scale; ^dCRP: C-Reactive Protein; ^eMI: Myocardial Infarction; ^fAKI: Acute Kidney Injury, ^gLOS: Length of Stay, ^hIQR: Interquartile Range; ⁱSD: Standard Deviation.

Univariable analysis identified the following factors to be significantly associated with 30-day mortality: age, CCI, systolic blood pressure, logCRP, logCreatinine, blood transfusion, osteopenia, and the surgical indication being ischaemia (Table 2). Furthermore, multivariable analysis with backward stepwise selection identified the following variables as significantly associated with 30-day mortality: age (IRR=1.05, 95% CI: 1.02, 1.08), logMortality P-POSSUM (IRR=1.58, 95% CI: 1.16, 2.15), ischaemia as the indication for surgery (IRR=1.94, 95% CI 1.05, 3.57), PRB transfusion (IRR=1.66, 95% CI: 1.13, 2.42), and cancer (IRR=1.57, 95% CI: 1.01, 2.44) (Table 3).

Table 2. Incidence rate ratios of univariable analysis using Poisson regression

^aIRR: Incidence Rate Ratio, ^bCI: Confidence Interval, ^cBMI: Body Mass Index, ^dCCI: Charlson’s Comorbidity Index, ^eCRP: C-Reactive Protein, ^fPRB: Packed Red Blood cells; ^gASA: American Society of Anaesthesiologists.

Table 3. Incidence rate ratios of multivariable analysis with backward stepwise selection using Poisson regression

^aIRR: incidence rate ratio; ^bCI: confidence interval; ^cASA: American Society of Anaesthesiologists; ^dPRB: Packed Red Blood.

Variables demonstrating significant associations with mortality were entered into prognostic value assessment. Mortality P-POSSUM, morbidity P-POSSUM and osteopenia on their own (models 3, 4, 5 in Figure 1 and Table 4), all demonstrated poor prognostic value for 30-day mortality with area under ROC curve (AUC) of 0.49, 0.47, and 0.59 respectively (Table 4, Figure 1). Additionally, the prognostic tests either demonstrated poor sensitivity (mortality P-POSSUM and morbidity P-POSSUM) or both poor sensitivity and specificity (osteopenia). When adjusted for other statistically significant covariates, the models with mortality P-POSSUM, morbidity P-POSSUM or both showed much improved prognostic value with AUC >0.80 and both sensitivity and specificity >75% (Table 4, Figure 1). Further, there was a statistically significant improvement in the performance of multivariable Poisson regression models including important covariates compared to mortality P-POSSUM, morbidity P-POSSUM, or osteopenia alone (c² = 77.00, p <0.001) (Table 4).

Table 4. Comparison of areas under ROC^b curves

^aAt optimum cut-off point based on Youden’s J statistic; ^bROC: Receiver operating characteristic, ^cAUC: Area under ROC curve, ^dSE: Standard error, CI: Confidence interval

Model 1 covariates: ASA, age, logMorbidity POSSUM, logMortality POSSUM, surgery indication, PRB transfusion, cancer, systolic blood pressure, surgical procedure

Model 2 covariates: ASA, age, logMortality POSSUM, surgery indication, PRB transfusion, cancer, systolic blood pressure, surgical procedure

Model 3 covariates: ASA, age, logMorbidity POSSUM, surgery indication, PRB transfusion, cancer, systolic blood pressure, surgical procedure

Model 4 covariate: logMortality POSSUM

Model 5 covariate: logMorbidity POSSUM

Model 6 covariate: Osteopenia

Fig. 1. Area under ROC curves.

Discussion

The ability to reliably predict a patient’s survival is of great relevance in the emergency surgical setting to facilitate appropriate care. Several previous studies have found that prognostic factors such as general frailty and the development of severe post-operative complications are of greatest importance in affecting clinical outcomes [17,18]. In order to avoid age discrimination and to guide appropriate clinical decision-making, predictive tools have been developed. While P-POSSUM is a widely used system, it is a general surgical predictive tool which is not specifically aimed at the geriatric population. The predictive tool has been criticised for overestimating expected mortality and for not taking aspects such as surgeon-related variables into account [19,20]. Alternative systems such as the measurement of frailty have therefore received focus. There is, however, currently no consensus as to which system to use in the elderly.

Thirty-day mortality rate in the current study was 31.1% reflecting the high mortality risk following emergency laparotomy in the elderly population. This figure is in line with previous mortality rates for elderly patients following abdominal surgery: when considering patients aged 80 and above Clarke et al., [21] demonstrated 38% mortality (compared to the overall mortality rate of 20%) and Saunders et al. demonstrated a death rate of 24.4% [7]. Diagnostic test analysis demonstrate that mortality P-POSSUM show weak predictive value in the current geriatric patients with an AUC of only 0.49. Additionally, the discriminatory ability of mortality P-POSSUM to identify those who died as deceased was very poor with a sensitivity score of 0.11.  Osteopenia did demonstrate a slight improvement with moderate prognostic value with an AUC of 0.59 and sensitivity of 0.60. The current study design is unable to fully offer an account for this difference, but one possible explanation could be the fact that P-POSSUM does not take frailty into account whereas osteopenia is, by many, considered to be a marker of general frailty.

Further complexity is added to the subject of frailty by the fact that no generally accepted definition for the term exists. The concept does, however, commonly include some combination of weight loss, muscle wasting, loss in grip-strength and limited physical activity in a typical day [22]. The presence of frailty in elderly surgical patients has previously been emphasised and linked to poorer clinical outcomes [22]. The development of a frailty index for geriatric patients undergoing emergency surgery has been explored with some success. Joseph and colleagues found that the frailty index demonstrated a strong prediction for developing major complications and did this independently of age and ASA classification [23].

Furthermore, our results suggest that mortality P-POSSUM requires the additional information of ASA classification, age, surgical indication and procedure as well as PRB transfusion requirements, systolic blood pressure on admission and information on any concurrent cancer diagnosis in order to achieve a strong predictive power (AUC = 0.83, sensitivity and specificity >0.75). This finding adds to the discussion of complexity when it comes to reliably predicting mortality amongst the elderly. However, it should be pointed out that these are all variables which are available directly at admission and it is therefore feasible to apply pre-operatively. Some of these variables are already taken into consideration in P-PUSSOM calculation (age, systolic blood pressure, and malignancy status), but how the additional variables (ASA classification, indication of surgery and type of procedure, and blood transfusion) would be combined with or into P-POSSUM in a mortality prediction calculator remains to be answered. A possible alternative is to assess whether a recalibration of P-POSSUM in the geriatric population would result in sufficiently improved predictive power.

To the best of our knowledge previous comparative validations of P-POSSUM and osteopenia are scant in the literature. However, the current study has multiple limitations inherent to its retrospective nature and limited sample size. Firstly, sarcopenia had to be excluded as a prognostic indicator due to the lack of patients in the study population meeting CT criteria for sarcopenia. Secondly, due to the lack of information on performance status prior to hospital admission we were unable to compare the prognostic value of P-POSSUM and osteopenia to a frailty index.

In the current investigation, P-POSSUM and osteopenia alone proved to be poor predictors of 30-day mortality in geriatric patients following emergency laparotomy. P-POSSUM adjusted for other patient covariates improve outcome prediction.

Conflict of Interest: The authors declare that they have no conflict of interest and have no funding sources to declare.