The Development, Internal and External Validation of a Circumcision Complications Risk Calculator for an African Population: Prevention of Circumcision Complications via Pre-circumcision Complication Risk Profiling in Ghana

Abstract

Background and objective Circumcision complications from clinical and non-clinical procedures pose significant health risks in Ghana. In light of this, tools that predict and prevent these mishaps using individual sociodemographic risk factors as digital biomarkers are urgently needed. Mobile health (mHealth) technologies offer a promising platform for improving circumcision safety through digital risk profiling. In this study, we aimed to develop a mobile app-based digital risk calculator for preventing circumcision complications in Ghana by leveraging digital biomarkers and risk profiling. Methods We conducted a five-year retrospective analysis of hospital-based data involving a total of 217 participants (186 for model development and 31 for external validation), identifying key risk factors including demographics, circumciser skill level, and provider facility type. Embedded, but not explicit, was the circumcision-seeking behavior of participants and thus, the geospatial distribution of complications. These variables were integrated into a logistic regression model. Internal and external validation of the model was conducted. The model was then deployed via an "R: A Language and Environment for Statistical Computing" platform, embedded into a mobile app designed for healthcare providers and parents. Pilot testing assessed app usability in 30 adult participants. Results The app categorized patients into low, moderate, and high-risk groups. The diagnostic model achieved a specificity of 96.08%, a positive predictive value (PPV) of 64.71%, and a negative predictive value (NPV) of 86.98%, correctly classifying 84.95% of cases. Sensitivity was 33.33%. The Hosmer-Lemeshow goodness-of-fit test (χ2 = 11.05, p = 0.199) confirmed the model fit. The receiver operating characteristic (ROC) analysis showed excellent discrimination [area under the curve (AUC) = 0.8895]. External validation and usability testing yielded favorable results. Conclusions This mobile app offers a valuable tool for real-time circumcision risk assessment, enhancing safety outcomes. Future research should aim to incorporate machine learning to optimize predictive performance.

Keywords: circumcision in males; ghana; logistic regression models; prevention and control of circumcision complications; risk assessment; sociodemographic characteristics as digital biomarkers; validation study; web/mobile applications.

Figure 1. Conceptual framework diagram for the study

Conceptual framework illustrating the factors influencing circumcision outcomes This diagram outlines the key variables and hypothesized relationships that affect circumcision outcomes. Factors are categorized into patient-related, provider-related, procedural, and contextual domains. Arrows indicate the direction of influence, demonstrating how these elements interact to impact the overall outcome of circumcision procedures Source: authors’ original creation

Figure 2. Geospatial distribution map for circumcision cases and circumcision-related complications over the five-year study period

The map illustrates spatial clustering and communal patterns of both primary circumcision procedures and reported complications of circumcisions across the study area

Figure 3. Receiver-operator characteristic curve for the model equation (AUC = 0.78)

Receiver operating characteristic (ROC) curve illustrating the performance of the predictive model in distinguishing between circumcision cases with and without complications. The area under the curve (AUC) is 0.78, indicating good discriminative ability of the model. The curve demonstrates the trade-off between sensitivity and specificity across various threshold values. An AUC of 0.78 suggests that the model has a 78% chance of correctly differentiating a randomly chosen complication case from a non-complication case. This evaluation supports the model's clinical utility in risk stratification and decision-making

Figure 4. A sensitivity-specificity plot showing the Youden’s index cut-off point (decision point) = 0.200

Sensitivity-specificity plot illustrating the diagnostic performance of the predictive model across varying threshold values. The optimal decision threshold, determined using Youden’s index, is marked at 0.200. This cut-off point maximizes the combined sensitivity and specificity, thereby identifying the most effective point for classifying circumcision-related complications. The plot supports the evaluation of trade-offs between false positives and false negatives in clinical decision-making

Figure 5. The appearance of the interface of the digital mobile/web application

User interface of the digital mobile/web-based application developed for predicting the risk of circumcision-related complications. The interface features structured input fields for key predictors such as age, ethnicity, residence, parental occupation, and facility type. It displays real-time risk estimation outputs and decision-support messages to guide clinical or parental decision-making. Designed for usability across devices, the interface supports both mobile and desktop platforms to ensure accessibility for healthcare providers in diverse settings