Test for multi-collinearity.

Background
Prostate cancer is the most prevalent male malignancy in Ghana, with a high-risk of metastatic progression. Early detection and adequate disease severity stratification are crucial for timely intervention, comprehensive management, and improved outcomes. This study evaluates and compares the predictive abilities of a multiparametric model and a PSA-alone model in forecasting metastasis in prostate cancer patients.
Objective
To compare the performance of a multiparametric model and a PSA-alone model in predicting metastasis in prostate cancer patients in Ghana.
Methodology
Logistic regression analyses were conducted on a dataset of 426 prostate cancer cases. The multiparametric model included variables such as age, BMI, marital status, ethnicity, socioeconomic status, clinical stage by DRE findings, PSA levels, and Gleason score. The PSA-alone model focused solely on PSA levels. Model performance metrics included Pseudo R-Squared, AUC, sensitivity, specificity, accuracy, PPV, NPV, FPR, FNR, and F1-Score. The Hosmer-Lemeshow test assessed the goodness-of-fit for the multiparametric model. All analyses were conducted at a 5% level of significance.
Results
The multiparametric model achieved a Pseudo R-Squared of 71.17%, AUC of 97.18%, sensitivity of 93.20%, specificity of 96.21%, accuracy of 92.25%, PPV of 85.62%, NPV of 96.24%, FPR of 8.24%, FNR of 6.80%, and F1-Score of 81.02%. The Hosmer-Lemeshow test yielded a non-significant p-value of 0.2405. The PSA-alone model had sensitivity of 32.24%, specificity of 91.76%, accuracy of 88.03%, PPV of 77.47%, NPV of 92.02%, FPR of 3.79%, FNR of 67.76%, F1-Score of 45.76%, and AUC of 73.79%. The multiparametric model’s Prevalence Yield was 32.15% and Sensitivity Yield was 32.15%, compared to the PSA-alone model’s 6.95% and 13.32%, respectively.
Conclusion
Both models effectively predict metastasis in prostate cancer patients. The multiparametric model shows superior overall performance with higher Pseudo R-Squared, AUC, and a better balance in sensitivity, specificity, and accuracy. These results suggest the multiparametric model as a more robust tool for metastasis risk assessment in resource-poor settings. However, clinical context and patient characteristics should guide model choice for optimal outcomes.