Artificial Intelligence–Based Perioperative Risk Assessment for Elderly Patients Undergoing Inguinal Hernia Repair in a Resource‐Limited Setting

Abstract

Population aging has increased the number of older adults requiring surgical care, including in low- and middle-income countries (LMICs) [1]. Elderly patients often present with multimorbidity, reduced physiological reserve, and functional vulnerability, all of which influence postoperative outcomes. Inguinal hernia repair is a common procedure in this population and, although frequently considered low risk, outcomes among older adults are variable [2]. Routine perioperative risk assessment commonly relies on age and comorbidity burden, offering limited insight into frailty, particularly in resource-constrained settings [3]. Artificial intelligence (AI)–based prediction models may enhance risk stratification by integrating multiple patient-level variables [4]. Evidence supporting their use in routine general surgery in sub-Saharan Africa remains limited. This study evaluated an AI-based perioperative risk prediction model for elderly patients undergoing inguinal hernia repair at a Nigerian teaching hospital and compared its performance with conventional clinical risk assessment. A retrospective cohort analysis was performed involving adults aged 65 years or older who underwent inguinal hernia repair at a tertiary teaching hospital in Nigeria. Eligible cases included both elective and emergency procedures. Clinical information was retrieved from perioperative records, including ward notes, anesthesia documentation, operative reports, and discharge summaries, using a predefined data extraction tool. All patients underwent open anterior inguinal hernia repair, most commonly using a Lichtenstein mesh technique for elective cases, with tissue repair reserved for selected emergency or contaminated procedures; perioperative antibiotic prophylaxis was routinely administered before skin incision. Recorded patient characteristics comprised age, sex, and surgical urgency. Pre-existing medical conditions documented at the time of surgery included systemic hypertension, diabetes mellitus, chronic pulmonary disease, chronic kidney disease, ischemic heart disease or cardiac failure, and prior cerebrovascular events. Patients with two or more chronic illnesses were categorized as having multimorbidity. Because formal geriatric assessment tools were not routinely available, vulnerability was evaluated using clinically observable frailty markers. These included limited ambulation or use of mobility aids, dependence in basic activities of daily living, history of recurrent falls, documented cognitive impairment, low body mass index or unintentional weight loss, and preoperative anemia (hemoglobin < 10 g/dL). The presence of at least one marker was considered indicative of frailty. Postoperative outcomes evaluated were in-hospital complications, length of postoperative hospital stay, and new functional dependence at discharge. The overall median postoperative length of stay was 3 days. Prolonged hospitalization was predefined as a stay exceeding 5 days and occurred in 26.7% of patients. A supervised artificial intelligence–based model incorporating demographic, comorbidity, frailty, and operative variables was developed, with predictive performance assessed using receiver operating characteristic analysis. Conventional clinical risk assessment reflected routine preoperative surgical evaluation in our setting and was based on chronological age, documented comorbidities, operative urgency (elective vs. emergency), and overall clinician judgment, without the use of formal geriatric assessment tools, validated frailty indices, or algorithmic risk scores. Predictive performance of this traditional assessment was evaluated using the same outcome measures and receiver operating characteristic analysis applied to the artificial intelligence–based model. Ethical approval was obtained from the institutional research ethics committee prior to data extraction. Sixty elderly patients were included in the analysis, with a mean age of 71.4 ± 5.8 years; males constituted 90.0% of the cohort. Hypertension and diabetes mellitus were the most frequently documented comorbidities, and 41.7% of patients had two or more chronic medical conditions. Features suggestive of frailty were identified in 35.0% of cases. Emergency procedures accounted for one-fifth of all operations. Fourteen patients (23.3%) developed at least one postoperative complication, and several individuals experienced more than one complication during the index admission. Most common post-operative complication is surgical site infection (Table 1). There was no perioperative mortality. Prolonged hospitalization and deterioration in functional status were closely linked to the occurrence of postoperative complications. Among patients who developed complications, 85.7% had a postoperative length of stay exceeding 5 days, compared with 8.7% of those without complications. Likewise, new-onset functional dependence occurred predominantly in patients with postoperative complications, indicating a strong association between morbidity and adverse recovery trajectories (Table 2). The AI-based model demonstrated improved predictive accuracy for postoperative complications (AUC = 0.81) compared with traditional clinical assessment (AUC = 0.66). Increasing AI-predicted risk was associated with progressively higher complication rates. Emergency surgery, frailty indicators, multiple comorbidities, and elevated AI risk scores were independently associated with adverse postoperative outcomes (Table 3). This study shows that AI-based perioperative risk prediction enhanced identification of elderly patients at risk of adverse outcomes following inguinal hernia repair in a low-resource setting. The observed complication burden is consistent with reports from geriatric surgical populations, underscoring the influence of physiological vulnerability rather than age alone on postoperative recovery [5]. Importantly, the superior performance of the AI model compared with conventional assessment highlights its capacity to account for complex interactions between clinical and functional variables [6]. The postoperative complication rate observed in this cohort (23.3%) is higher than rates typically reported for inguinal hernia repair in high-income Western settings, where morbidity following elective repair is generally low [2]. This disparity likely reflects important differences in patient selection, baseline health status, timing of presentation, and availability of perioperative resources across health systems [3, 7]. In the present study, a substantial proportion of patients had multimorbidity, frailty indicators, and emergency presentations, all of which are well-established contributors to increased postoperative risk in older surgical populations [4, 8]. In contrast, Western series commonly involve younger, fitter patients undergoing elective repair within structured and well-resourced perioperative care pathways, resulting in lower reported morbidity [2, 7]. Therefore, direct numerical comparison of complication rates across these settings is inappropriate, and the observed morbidity should be interpreted as representative of surgical care in resource-limited environments rather than an indicator of procedural risk alone [1, 3]. The duration of hospitalization observed in this study is longer than that reported in many Western settings, where elective inguinal hernia repair is frequently performed as a day-case or outpatient procedure [2]. This difference reflects structural and health-system factors rather than surgical complexity alone. In resource-limited environments, routine same-day discharge is often constrained by delayed presentation, higher prevalence of comorbidity and frailty, limited access to perioperative optimization, and reduced availability of post-discharge support systems [3, 4, 7]. Additionally, inpatient observation is commonly used to mitigate risks related to pain control, wound monitoring, and early postoperative complications in the absence of robust ambulatory follow-up pathways [4, 7]. Consequently, longer hospital stays in this context represent a risk-mitigation strategy rather than inefficiency, and comparisons with outpatient models in high-income settings should be interpreted cautiously [1, 3]. Frailty emerged as a key determinant of postoperative outcomes, reinforcing its importance as a marker of diminished physiological reserve and impaired recovery potential [9]. The ability of the AI model to anticipate loss of functional independence is particularly relevant in elderly care, where preservation of autonomy is a central outcome of interest. In environments with limited specialist resources, AI-driven tools may provide practical decision support by enhancing risk stratification using routinely collected clinical data. The findings of this study should be interpreted within their demographic and health-system context. The median age of the cohort was 71 years, which reflects commonly applied definitions of older adults in low- and middle-income countries but may not align with perceptions of “elderly” in many Western populations, where surgical candidates often present at more advanced ages [1, 3]. Differences in baseline life expectancy, comorbidity burden, functional reserve, and access to perioperative optimization further limit direct extrapolation of these results to high-income settings [3, 4, 7, 10]. Accordingly, the present findings are most applicable to similar resource-limited environments and should be viewed as context-specific rather than universally generalizable [1, 3]. While the single-center design and modest sample size limit generalizability, the findings offer important preliminary evidence supporting the feasibility of AI-guided surgical risk assessment in sub-Saharan Africa. Further prospective, multicenter studies are needed to validate these findings and explore integration into routine clinical workflows. Opeyemi Qozeem Asafa: conceptualization, formal analysis, investigation, methodology, project administration, resources, software, supervision, writing – original draft, writing – review and editing. Aishat Omowunmi Asafa: data curation, software, writing – review and editing, writing – original draft, project administration. Emmanuel Oladayo Folami: writing – review and editing, writing – original draft. Kehinde Adesola Alatishe: writing – review and editing, project administration, writing – original draft, supervision. Olumuyiwa Tope Ajayeoba: writing – original draft, writing – review and editing. Kehinde Awodele: writing – review and editing, project administration, writing – original draft, supervision. Ismail Idowu Uthman: software, supervision, writing – original draft, writing – review and editing. Oyelami Naheem Kehinde: writing – original draft, writing – review and editing. Mustapha Babatunde: project administration, supervision, resources, writing – review and editing, writing – original draft. Oyeniyi Ganiyu Adebukola: writing – original draft, writing – review and editings. The authors have nothing to report. This study was conducted in accordance with the Declaration of Helsinki. Ethical approval was obtained from the Osun State University Teaching Hospital Health Research Ethics Committee. Individual informed consent was waived due to the retrospective design. All patient data were anonymized and handled with strict confidentiality. The authors declare no conflicts of interest. The data that support the findings of this study are available from the corresponding author upon reasonable request.

Ähnliche Arbeiten