Incidence of postoperative mortality and validation of the ‘Preoperative Score to Predict Postoperative Mortality’ (POSPOM) in Thailand

Abstract

Objectives

To determine the incidence of in-hospital postoperative mortality and validate the Preoperative Score to Predict Postoperative Mortality (POSPOM).

Design

Retrospective case–control study.

Setting

A tertiary university hospital in Thailand.

Participants

All patients who underwent surgical procedures in 2019 under any anaesthesia technique.

Main outcome measures

Incidence of in-hospital postoperative mortality and POSPOM score performance (sensitivity, specificity and predictive values).

Results

A total of 39 674 patients underwent surgery in 2019. The in-hospital postoperative mortality rate was 0.76% (95% CI 0.68% to 0.85%). After excluding cases per POSPOM criteria, 270 deceased patients remained. The POSPOM score was validated against these 270 cases and 270 randomly selected survivors, yielding an area under the receiver operating characteristic curve of 0.73 (95% CI 0.69 to 0.77). At a cut-off of 21, the sensitivity was 71.5% (95% CI 65.7% to 76.8%), and the specificity was 61.9% (95% CI 55.8% to 67.7%).

Conclusions

This single-centre study found a low in-hospital postoperative mortality rate. The POSPOM score demonstrated moderate predictive accuracy for postoperative mortality risk and may be useful for preoperative risk stratification.

Trial registration number

https://www.thaiclinicaltrials.org/show/TCTR20200330001.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• First validation of Preoperative Score to Predict Postoperative Mortality in Asia, whereas prior investigations were conducted mainly in Europe.

• Single-centre design may limit external validity.

• Manual, retrospective data collection could result in missing or incomplete information.

Background

In-hospital postoperative mortality is widely regarded as a key indicator of surgical quality and patient safety, as shown by findings from the International Surgical Outcomes Study.¹ Mortality rates are influenced by the demographic characteristics of the study population, the interval after surgery, the hospital’s level of care, and the country’s economic status (table 1). These rates vary widely, ranging from 0.08% to 21.5%.1,18 However, data encompassing all surgeries and age groups in Thailand are outdated.

Table 1. Incidences of postoperative mortality in published studies.

Reference	Country	Type of study	Patient characteristics	N	Incidence (%)	Duration after surgery	Study period
Fecho et al2	USA (North Carolina)	Retrospective cohort (single tertiary hospital)	All surgeries, all age	12 739	0.57 2.1	48 hours 30 days	April 2004–March 2005
Noordzij et al3	Netherlands	Retrospective cohort (national database)	Age over 20, elective, inpatient, open surgery	3 667 875	1.85	30 days	January 1991– December 2005
Pearse et al4	European countries	Prospective cohort (multicountry)	Age over 16, inpatient, non-cardiac surgery	46 539	4 (varied 1.2–21.5)	NA	4–11 April 2011
International Surgical Outcomes Study group1	International	Prospective cohort (multicountry)	Age above 18, inpatient, elective surgery	44 814	0.5	NA	April–August 2014
Smith et al5	USA	Retrospective cohort (multiple veteran hospitals)	Veterans, all surgeries	212 733	1.03	365 days	October 2011–September 2013
Le Manach et al6	France	Retrospective cohort (national database)	Age over 18, excluding obstetric and monitored care anaesthesia	2 717 902	0.47	In-hospital	January 2010– December 2010
Spence et al7	International	Prospective cohort (multicountry: VISION Study)	Age over 45, inpatient, non-cardiac surgery	40 004	1.26	30 days	August 2007– November 2013
Kaiser et al8	USA	Retrospective cohort (multicenter)	Age over 18, non-cardiac surgery	1 859 917	1.25	30 days	January 2008– December 2012
Liew et al9	Singapore	Retrospective cohort (single tertiary hospital)	Age over 45, non-cardiac surgery	2163	5.90	1 year	January–July 2015
Gurney et al10	New Zealand	Retrospective cohort (national database)	All age, any surgery with general anaesthesia	1 836 683	0.5 0.9	30 days 90 days	January 2005– December 2017
Layer et al11	Germany	Retrospective cohort (single tertiary hospital)	Age over 18, excluding procedures without relevant anaesthesia	199 780	2.03	In-hospital	January 2006– December 2017
Madsen et al12	USA	Retrospective cohort (multicenter)	Adults, non-cardiac surgery	5 777 108	0.95	30 days	2012–2018
Charuluxananan et al13	Thailand	Prospective cohort (multicenter)	All age, all surgeries	163 403	0.28	24 hours	February 2003–January 2004
Charuluxananan et al14	Thailand	Prospective cohort (multicenter)	All age, all surgeries	333 219	0.13	24 hours	January–December 2015
Siriphuwanun et al15	Thailand (Chiang Mai)	Retrospective cohort (single tertiary hospital)	All age, emergency surgery	46 624	1.36	24 hours	2003–2011
Wirachpisit et al16	Thailand (Bangkok)	Retrospective cohort (single tertiary hospital)	All age, emergency surgery	3250	0.68	48 hours	June 2014–May 2015
Angkasuvan and Hombut17	Thailand (Songkla)	Retrospective cohort (single tertiary hospital)	All age, all surgeries	16 000	0.12	24 hours	2007–2016
Chomchoey and Thawitsri18	Thailand (Bangkok)	Retrospective cohort (single tertiary hospital)	Age over 18, all surgeries with anaesthesia	34 590	0.08	30 days	September 2018–August 2020

NA, not available.

Several tools have been developed to assess postoperative complications, including death, with the aim of refining perioperative management and improving individual prognoses. One such instrument, the Preoperative Score to Predict Postoperative Mortality (POSPOM), was introduced in 2016.⁶ This scoring system was created and validated using data from more than 5 million adult patients in France in 2010. It demonstrated strong predictive ability, with an area under the receiver operating characteristic curve (AuROC) measuring 0.929.⁶ The POSPOM score incorporates factors such as patient age, comorbidities and the nature of the surgical intervention. A patient’s total score then provides an estimated probability of in-hospital postoperative mortality. The POSPOM score has been repeatedly validated for diverse procedures and surgeries, showing a robust correlation with postoperative mortality.1119,23 Nevertheless, all validation studies have been conducted in European settings. Given the marked differences in patient populations, hospital infrastructures and cultural factors, the applicability of these results may not extend across continents.

We conducted a retrospective case–control analysis with the primary objective of reporting in-hospital postoperative mortality at a tertiary hospital in Thailand. As a secondary objective, we applied the POSPOM score to patients who died postoperatively and compared that score with randomly selected postoperative survivors in a case–control design. This evaluation was undertaken to determine the predictive accuracy of the POSPOM score.

Methods

Study design and participants

All data were fully anonymised before they were accessed, and the board waived the requirement for informed consent. Data for all operative cases in 2019 were obtained from both the Department of Anesthesiology and the Siriraj Data Management Unit (figure 1).

Figure 1. Study flow chart. GI, gastrointestinal; POSPOM, Preoperative Score to Predict Postoperative Mortality.

The study population included patients who underwent surgery under anaesthesia between 1 January 2019 and 31 December 2019. We identified patients who died during the same admission in which their surgery took place, excluding any patients with missing data. This subset was used to calculate the incidence of in-hospital postoperative mortality.

POSPOM excludes three patient groups⁶: those under 18 years of age, those undergoing obstetric procedures and those who had procedures without anaesthesia. Examples of procedures without anaesthesia include cardiac catheterisation, endoscopy and certain ophthalmic interventions. Therefore, these patient groups were excluded from our study. We then randomly selected surviving patients at a 1:1 ratio relative to the mortality group and applied POSPOM to evaluate its predictive accuracy.

Data collection and outcome measures

We systematically recorded demographic data (age and sex), American Society of Anesthesiologists (ASA) physical status classification, comorbidities, type of surgery, emergency status and the interval from surgery to death. We also documented the anaesthesia technique (general or non-general), preoperative comorbidities, postoperative destination (postanaesthetic care unit, high-dependency unit or intensive care unit (ICU)), extubation status and the time from surgery to death. If a patient underwent multiple operations during a single admission, we selected the surgery that produced the highest POSPOM score for data collection.

Preoperative Score to Predict Postoperative Mortality

We assessed the accuracy of POSPOM using the method described by Le Manach et al.⁶ This scoring system assigns points according to age, surgical category and up to 15 comorbidities, which together can contribute a maximum of 32 points. The total POSPOM score ranges from 0 to 70, and higher scores indicate greater risk of postoperative mortality. A detailed explanation of POSPOM variables is provided in online supplemental appendix 1.

Sample size calculation

The sample size calculation was based on an approximate 2% 30-day postoperative mortality rate reported in previous studies.² With an acceptable margin of error of 0.15% (two-sided, 95% CI), the required sample size was 33 465. We anticipated that 1 year of surgical data would encompass enough deceased cases to address our primary objective. For our secondary objective—validating the discriminatory ability of POSPOM—we conducted a 1:1 case–control study. An equal number of surviving postoperative patients from 2019 were randomly selected to match the number of deceased cases.

Statistical methods

All statistical analyses were carried out using PASW Statistics, V.18 (SPSS Inc, Chicago, IL, USA). The Shapiro-Wilk test was used to assess data distribution, with significance set at p<0.05. Continuous variables are reported as means with SD for normally distributed data or as medians with IQRs for non-normally distributed data. In-hospital mortality rates are presented as percentages with 95% CIs. We used the χ²or Fisher’s exact test to compare categorical variables between deceased and surviving participants, and the Mann-Whitney U test for continuous variables. A p<0.05 indicated statistical significance. The OR with a 95% CI was also calculated to identify contributing factors.

We calculated the POSPOM score and evaluated its accuracy by assessing sensitivity, specificity, and positive and negative predictive values. We then determined the AuROC to confirm the predictive capability of POSPOM in differentiating between patients who died postoperatively and those who survived.

Results

The study flow chart is shown in figure 1. In 2019, 39 674 patients underwent surgery. Among these, 742 patients died, with 305 of those dying during the same admission in which anaesthesia was provided. Two patients were excluded because of invalid data, leaving 303 in-hospital postoperative deaths. This number corresponds to an incidence of 0.76% (95% CI 0.68% to 0.85%). When stratified by time to postoperative death, mortality within 24 hours, 72 hours, 7 days, 14 days and 30 days was 0.03% (95% CI 0.03% to 0.03%), 0.15% (95% CI 0.15% to 0.15%), 0.24% (95% CI 0.24% to 0.25%), 0.38% (95% CI 0.37% to 0.38%) and 0.55% (95% CI 0.55% to 0.56%), respectively.

Table 2 presents demographic data for the deceased participants. Their median age was 68.2 years (range: 11 days to 96 years). 90% had an ASA classification of III or higher, and 85.1% underwent general anaesthesia.

Table 2. Demographic data for in-hospital postoperative deceased patients.

	Frequency (%), median (IQR) (n=303)
Male sex	167 (55.3)
Age; median (IQR)	68.21 (53.69–80.05)
ASA classification
I	1 (0.3)
II	29 (9.6)
III	151 (49.8)
IV	103 (34.0)
V	19 (6.3)
Emergency	157 (51.8)
General anaesthesia	258 (85.1)
Postoperative extubation	98 (32.3)
Postoperative dispatch
Post anaesthesia care unit	87 (28.7)
High dependency unit	42 (13.9)
Intensive care unit	174 (57.4)
Time interval from surgery to death
<24 hours	28 (9.2)
24–72 hours	31 (10.2)
72 hours–7 days	37 (12.2)
7–30 days	123 (40.6)
>30 days	84 (27.7)

ASA, American Society of Anesthesiologists.

Demographic data comparison based on POSPOM

Table 3 compares demographic characteristics between deceased (n=270) and surviving postoperative patients. The deceased group had a significantly higher median age than the survivors. Based on the POSPOM criteria, the two cohorts differed significantly in nine factors. The exceptions were cardiac arrhythmia, chronic obstructive pulmonary disease, peripheral arterial disease, organ transplantation, chronic alcohol abuse and hemiplegia/paraplegia. Surgeries with notably higher mortality included cardiac procedures, major gastrointestinal operations, digestive endoscopy, multiple trauma-related surgery and other miscellaneous interventions. The median total POSPOM score was significantly higher among deceased patients (27.0, IQR 21.0–31.0) than survivors (20.0, IQR 15.0–24.0, p<0.001).

Table 3. Demographic data for patients evaluated using POSPOM Score criteria.

	Deceased (n=270)	Survivors (n=270)	P value	OR (95% CI)
Age (years); median (IQR)	69.8 (59.3–80.4)	58.9 (41.8–70.1)	<0.001	1.0 (1.0 to 1.1)
Male sex; n (%)	151 (55.9)	121 (44.8)	0.010	1.6 (1.1 to 2.2)
Comorbidities
Ischaemic heart disease	68 (25.2)	24 (8.9)	<0.001	3.5 (2.1 to 5.7)
Cardiac arrhythmia	18 (6.7)	10 (3.7)	0.118	1.9 (0.8 to 4.1)
Cerebrovascular disease	43 (15.9)	18 (6.7)	0.001	2.7 (1.5 to 4.7)
Chronic obstructive pulmonary disease	9 (3.3)	3 (1.1)	0.080	3.1 (0.8 to 11.5)
Diabetic mellitus	89 (33.0)	49 (18.1)	<0.001	2.2 (1.5 to 3.3)
Peripheral arterial disease	19 (7.0)	10 (3.7)	0.086	2.0 (0.9 to 4.3)
Preoperative chronic dialysis	31 (11.5)	5 (1.9)	<0.001	6.9 (2.6 to 18.0)
Chronic renal failure	89 (33.0)	29 (10.7)	<0.001	4.1 (2.6 to 6.5)
Dementia	18 (6.7)	4 (1.5)	0.002	2.2 (1.3 to 3.8)
Transplant organs	2 (0.7)	1 (0.4)	1.000	2.0 (0.2 to 22.3)
Chronic respiratory failure	22 (8.1)	0 (0.0)	<0.001	NA
Active cancer	102 (37.8)	61 (22.6)	<0.001	2.1 (1.4 to 3.0)
Chronic alcohol abuse	16 (6.1)	13 (4.8)	0.518	1.3 (0.6 to 2.7)
Chronic heart failure or cardiomyopathy	43 (16.0)	11 (4.1)	<0.001	4.5 (2.3 to 8.9)
Hemiplegia or paraplegia or paralytic syndrome	12 (4.4)	5 (1.9)	0.085	2.5 (0.9 to 7.1)
Surgery types; n (%)
Cardiac surgery	32 (11.9)	9 (3.3)	<0.001	3.9 (1.8 to 8.3)
Interventional cardiorhythmology	10 (3.7)	5 (1.9)	0.190	2.0 (0.7 to 6.1)
Thoracic surgery	3 (1.1)	4 (1.5)	1.000	0.8 (0.2 to 3.4)
Neurosurgery	24 (8.9)	18 (6.7)	0.335	1.4 (0.7 to 2.6)
Interventional neuroradiology	6 (2.2)	3 (1.1)	0.504	2.0 (0.5 to 8.2)
Ear, nose and throat	25 (9.3)	30 (11.1)	0.477	0.8 (0.5 to 1.4)
Plastic and reconstructive surgery	2 (0.7)	9 (3.3)	0.033	0.2 (0.1 to 1.0)
Major digestive surgery	27 (10.0)	7 (2.6)	<0.001	4.2 (1.8 to 9.8)
Minor digestive surgery	18 (6.7)	29 (10.7)	0.093	0.6 (0.3 to 1.1)
Major vascular surgery	2 (0.7)	8 (3.0)	0.055	4.1 (0.9 to 19.5)
Minor vascular surgery	10 (3.7)	6 (2.2)	0.310	1.7 (0.6 to 4.7)
Major liver surgery	2 (0.7)	7 (2.6)	0.176	0.3 (0.1 to 1.4)
Minor liver surgery	1 (0.4)	7 (2.6)	0.068	0.1 (0.0 to 1.1)
Transplant surgery	2 (0.7)	0 (0.0)	0.499	NA
Renal transplantation	0 (0.0)	0 (0.0)	–	NA
Major urologic surgery	3 (1.1)	3 (1.1)	1.000	1.0 (0.2 to 5.0)
Minor urologic surgery	3 (1.1)	19 (7.0)	<0.001	0.5 (0.0 to 0.5)
Gynaecologic surgery	4 (1.5)	44 (16.3)	<0.001	0.1 (0.0 to 0.2)
Trauma-related orthopaedic surgery	0 (0.0)	9 (3.3)	0.004	NA
Minor orthopaedic surgery	12 (4.4)	13 (4.8)	0.838	0.9 (0.4 to 2.1)
Arthroplasty and spine surgery	3 (1.1)	21 (7.8)	<0.001	0.1 (0.0 to 0.5)
Ophthalmologic surgery	0 (0.0)	10 (3.7)	0.001	NA
Multiple trauma-related surgery	7 (2.6)	0 (0.0)	0.015	NA
Digestive endoscopy	48 (17.8)	6 (2.2)	<0.001	9.5 (4.0 to 22.7)
Other surgery	20 (7.4)	9 (3.3)	0.036	2.3 (1.0 to 5.2)

Bold values indicate statistical significance at p < 0.05, determined using the Mann-Whitney U test and the χ2 test.

NA, not available; POSPOM, Preoperative Score to Predict Postoperative Mortality.

Following univariate analysis and Bonferroni correction, 37.8% of the deceased group had an ASA classification above III, compared with 1.5% among survivors (p<0.001). The deceased group also showed higher proportions of emergency surgery (52.2% vs 12.2%, p<0.001) and general anaesthesia (85.9% vs 76.3%, p=0.004). In addition, 67.4% of deceased patients were transferred postoperatively to the ICU or intermediate wards, compared with 14.1% of survivors (p<0.001).

POSPOM score validation

After applying the POSPOM exclusion criteria, 270 valid deceased cases were included in the analysis. Their POSPOM scores were compared with those of 270 randomly selected living patients. Figure 2 shows a histogram of the score distributions for in-hospital postoperative deceased and surviving patients. The median score was significantly higher in the deceased group (27.0, IQR 21.0–31.0) than in the surviving group (20.0, IQR 15.0–24.0, p<0.001).

Figure 2. POSPOM score histogram for in-hospital postoperative deceased versus alive patients. The red dot represents the cut-point suggestion at POSPOM score over 21 which would mark the most balanced sensitivity and specificity. POSPOM, Preoperative Score to Predict Postoperative Mortality.

The receiver operating characteristic curve, depicted in figure 3, yielded an AuROC of 0.73 (95% CI 0.69 to 0.77).

Figure 3. Receiver operating characteristic (ROC) curve for POSPOM Score in the Thai population. AUROC, area under the receiver operating characteristic curve; POSPOM, Preoperative Score to Predict Postoperative Mortality.

As shown in table 4, the most balanced sensitivity and specificity were observed at a POSPOM cut-off above 21. At this threshold, sensitivity was 71.5% (95% CI 65.7% to 76.8%), and specificity was 61.9% (95% CI 55.8% to 67.7%). Given the in-hospital postoperative mortality rate of 0.76%, the positive predictive value was 1.41 (95% CI 0.41 to 3.52), and the negative predictive value was 99.6 (95% CI 97.8 to 100.0).

Table 4. Evaluation of individual properties within the POSPOM score in this study.

Score	N		Sensitivity % (95% CI)	Specificity % (95% CI)	PPV (95% CI)	NPV (95% CI)	Accuracy (95% CI)
20	204	124	75.6 (70.0 to 80.6)	54.1 (48.0 to 60.1)	0.01 (0.0 to 1.1)	100.0 (98.3 to 0.0)	54.1 (49.8 to 58.3)
	66	146
21	193	103	71.5 (65.7 to 76.8)	61.9 (55.8 to 67.7)	0.01 (0.0 to 1.3)	100.0 (98.5 to 0.0)	61.9 (51.6 to 65.0)
	77	167
22	185	91	68.5 (62.6 to 74.0)	66.3 (60.3 to 72.0)	0.02 (0.0 to 1.4)	100.0 (98.6 to 0)	66.3 (62.1 to 70.3)
	85	179
23	178	76	65.9 (59.9 to 71.6)	71.9 (66.1 to 77.1)	0.02 (0.0 to 1.5)	100.0 (98.7 to 0)	71.9 (67.9 to 75.6)
	92	194
24	168	66	62.2 (56.2 to 68.0)	75.6 (70.0 to 80.1)	0.02 (0.0 to 1.6)	100.0 (98.8 to 0)	75.6 (71.7 to 79.1)
	102	204
25	160	50	59.3 (53.1 to 65.2)	81.5 (76.3 to 85.9)	0.02 (0.0 to 1.8)	100.0 (98.9 to 0)	81.5 (77.9 to 84.7)
	110	220

Bold values indicate the selected POSPOM score of 21, which provides the most balanced sensitivity and specificity.

NPV, negative predictive value; POSPOM, preoperative score to predict postoperative mortality; PPV, positive predictive value.

Discussion

Our study offers valuable insights into the incidence of in-hospital postoperative mortality in the largest hospital in Thailand. We observed a mortality rate of 0.76% (95% CI 0.68% to 0.85%), which is relatively low. We also assessed the accuracy of POSPOM in predicting mortality, finding an AuROC of 0.73 (95% CI 0.69 to 0.77). The optimal cut-off score was 21, yielding a sensitivity of 71.5% (95% CI 65.7% to 76.8%) and a specificity of 61.9% (95% CI 55.8% to 67.7%).

Factors associated with mortality

Multiple factors influenced mortality risk. These were advanced age, ASA classification, type of anaesthesia, emergency status and specific comorbidities (ischaemic heart disease, cerebrovascular disease, diabetes mellitus, chronic renal failure, preoperative chronic dialysis, dementia, chronic respiratory failure, active cancer, chronic heart failure and cardiomyopathy). Certain procedures, such as cardiac surgery, major gastrointestinal surgery, digestive endoscopy and multiple trauma-related surgery, were also linked to higher mortality. Additionally, postoperative ICU admission emerged as a significant factor, consistent with a previous study.²⁴

Comparison with previous studies

Prior research has reported a wide range of postoperative mortality rates, often influenced by the defined interval from surgery until death (table 1). Many studies in countries outside Thailand use a 30-day time frame,^{2 3 7 8 10 12} whereas Thai investigations typically report only 24-hour mortality rates.13,1518 In our study, 30-day postoperative mortality was 0.55%, which is notably lower than rates from other settings. This difference may reflect our institution’s status as Thailand’s largest tertiary hospital and a key university centre, where advanced technology and abundant resources are available.

Additionally, Thai practice often involves hospitalising patients overnight before surgery. In contrast, Western healthcare systems frequently rely on ambulatory surgical approaches. This difference may result in a higher proportion of patients recorded as in-hospital, thereby lowering observed postoperative mortality rates. Our investigation found that Thailand had the most favourable outcome, with a 24-hour postoperative mortality of only 0.03%.

Earlier Thai studies on emergency surgery populations reported significantly higher postoperative mortality rates,^{15 16} suggesting that emergency procedures contribute substantially to mortality. At the 1-year mark, postoperative mortality in Singapore was 5.9%,⁹ whereas our rate was only 0.76%. This discrepancy may be explained by differences in methodology and patient populations. First, the Singaporean cohort focused on adults over 45 years, whereas we included patients aged 18 and older. Second, the Singaporean data encompassed all postoperative deaths, whereas we examined only in-hospital mortality within a single tertiary hospital. It is important to note that our hospital often frequently transfers stable patients to smaller hospitals for continued care, which may reduce the recorded mortality rate in our setting.

Validation of POSPOM score

The POSPOM scoring system was originally developed and validated in France, demonstrating excellent discrimination (AuROC 0.93).⁶ In our study, we evaluated this system by comparing POSPOM scores between deceased postoperative patients and a randomly selected cohort of survivors. The deceased group consistently showed higher total POSPOM scores, indicating that this tool can effectively predict postoperative mortality. We further confirmed its discriminative ability through an AuROC of 0.73, suggesting that POSPOM remains applicable in a Thai population despite differences in healthcare infrastructure and patient demographics compared with Western nations.

Validation in specific procedures and cohort studies

Several studies have validated POSPOM in specific surgeries. Two European investigations on radical cystectomy reported AuROC values of 0.86¹⁹ and 0.89,²⁰ while an AuROC of 0.82 was found in emergency abdominal operations²¹ and 0.71 in hip fracture surgery.²² Two recent cohort studies in Germany and the Netherlands evaluated entire surgical populations. The first, by Layer et al in Germany,¹¹ examined nearly 200 000 patients from 2006 to 2017, reporting a 2.0% mortality rate and an AuROC of 0.77. The authors concluded that POSPOM could be applied nationally with modifications to German coding. The second study, by Stolze et al in the Netherlands,²³ noted an AuROC of 0.82, reaffirming strong discrimination.

Compared with the three national cohort studies from France,⁶ Germany¹¹ and the Netherlands,²³ our study showed a slightly lower AuROC of 0.73. Although still fair in predictive capability, these differences may reflect variations in patient demographics, healthcare systems and data sources. In particular, delayed diagnosis of chronic conditions and strong family involvement in medical decisions are common in our setting. The French⁶ and Dutch²³ studies used large national databases, reporting AuROC values of 0.93 and 0.82. In contrast, our single-centre design resembles that of the German study, which showed an AuROC of 0.77,¹¹ suggesting that tertiary-hospital-based studies may yield more comparable results.

Interpretation of POSPOM score and misestimation of deaths

Although AuROC values for POSPOM appear promising, interpreting the score’s clinical utility requires caution. Some studies have highlighted a tendency to overestimate mortality. In one Belgian hip surgery investigation, the observed in-hospital mortality rate was 4.5%, whereas the POSPOM-predicted rate was 13.2%.²² Similarly, the previously mentioned Dutch study reported an observed rate of 0.5% yet a POSPOM-predicted rate of 1.3%.²³ In that study, a cut-off of 24 yielded 100% sensitivity but only 48% specificity, leading to a low positive predictive value of 1.1%.²³

Because our postoperative mortality rate was even lower, we recommend a POSPOM cut-off of 21. This threshold provides a more balanced sensitivity and specificity, offering greater clinical utility.

A German study reported an observed in-hospital mortality rate of 2.0%, only slightly higher than the expected rate of 1.7%.²⁴ The authors of that study noted that the quality and urgency of postoperative care also influence outcomes. Notably, patients who required unplanned ICU admission within 24 hours of surgery had a 13.5% mortality rate, despite low POSPOM scores.²⁴ This finding underlines the role of perioperative complications and failures in the rescue system, indicating that factors beyond preoperative risk scores significantly affect mortality.

Clinical factors associated with mortality

Our investigation identified several clinical factors associated with postoperative mortality: advanced age, ASA physical status, emergency status, type of anaesthesia and postoperative ICU admission. These findings mirror those of earlier studies.^{1 2 24} Notably, mortality risk can increase by up to 32.5-fold in patients who develop postoperative complications such as bleeding and sepsis.¹⁰ These observations underscore the need to enhance perioperative care for patients at heightened risk of complications.

Clinical implications

The POSPOM score, derived from a large cohort study and validated in diverse populations, has consistently shown relatively high AuROC values. It is suitable for adult patients receiving anaesthesia, with the exception of obstetric procedures and certain monitored anaesthesia care interventions (eg, cardiac catheterisation, gastrointestinal endoscopy and ophthalmologic surgery), which were not included in the original model.⁶ Although predictive accuracy may vary, POSPOM remains a valuable tool for estimating preoperative mortality risk and guiding clinical decisions. It can also be manually calculated, a practical feature in resource-limited settings or remote areas with limited access to electricity or the internet.

Alternative mortality risk calculators exist. One widely recognised example in Thailand is the Surgical Risk Calculator developed by the American College of Surgeons National Surgery Quality Improvement Programme.²⁵ This tool uses 21 variables to predict adverse postoperative outcomes, such as mortality, major complications and prolonged hospital stays. However, it does not provide an open-access predictive equation, preventing external validation at our institution.

Reilly et al conducted a comprehensive review of numerous mortality prediction tools. They identified only four tools with a high predictive accuracy after excluding those without external validation, those requiring unavailable preoperative data, or those demanding extensive patient information.²⁶ These instruments are the POSPOM scoring system, the Surgical Outcome Risk Tool,²⁷ the Surgical Mortality Probability Model²⁸ and the New Zealand Risk Predictor model.²⁹ An ideal instrument should predict mortality risk for both elective and emergency surgeries across different types of operations, yet remain easy to use. Online calculators are available for several tools; however, the POSPOM scoring tool requires 17 predictors and lacks an online option, which limits its immediate adoption. A more streamlined version may enhance its clinical utility.

The development of artificial intelligence further expands possibilities for predictive modelling. One study reported that an automated machine learning approach surpassed the American College of Surgeons National Surgery Quality Improvement Programme in predicting postoperative mortality.³⁰ Additional research is necessary to validate and refine these emerging methods, which could serve as more efficient and customisable preoperative prognostic tools across diverse patient populations.

Limitations and future directions

Several limitations should be acknowledged. First, our single-centre university hospital setting may reduce the generalisability of these findings to other institutions. Second, our data collection process was manual and retrospective, raising the possibility of missing information. Third, resource constraints led us to use a case–control study design. Although some research suggests that a well-designed case–control study can achieve discrimination comparable to a cohort study,³¹ a cohort study remains the standard for reliability. Moreover, the POSPOM score does not account for non-fatal postoperative complications, which also influence prognosis and management.

In the future, systematic data collection on national or global levels could broaden the application of POSPOM and enable meaningful international comparisons of postoperative morbidity and mortality. To our knowledge, this is the first validation of POSPOM in an Asian population. Prospective, multicentre research should be undertaken in settings with cultural and demographic characteristics distinct from Western populations. Ongoing efforts to develop and validate additional prediction tools will help ensure accurate evaluation of postoperative complications across diverse patient groups worldwide.

Conclusions

This study updates the incidence of postoperative mortality in Thailand’s largest tertiary hospital, indicating fewer deaths than previously reported. These findings provide a valuable benchmark for quality improvement initiatives. Moreover, POSPOM proved useful for identifying patients at higher risk of perioperative mortality.

The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript.

Data availability statement

Data are available on reasonable request.