Predicting Bone Metastasis in Prostate Cancer Patients Using Total Serum Prostate-Specific Antigen and Serum Alkaline Phosphatase: Model Development, Validation, and Deployment as a Digital Risk Estimator

Introduction This study addresses the global controversy over routine bone scans for newly diagnosed prostate cancer patients, focusing on the Ghanaian population. It aims to assess the predictive value of prostate-specific antigen (PSA) for bone metastasis and the role of serum alkaline phosphatase (ALP) in enhancing prediction. Methods This study was conducted at Korle Bu Teaching Hospital over 14 months and included 258 treatment-naïve prostate cancer patients. Clinical evaluation, PSA and ALP tests, and technetium-99 bone scans were performed. Chi-square and t-tests identified significant predictors of bone metastasis. The predictors were regressed logistically into a model, which was validated, trained, updated, and programmed into a digital risk calculator. All analysis was at a 0.05 significance level. Findings The mean age of participants was 68.18 ± 7.34 years (68.83 and 67.77 years for the metastatic and non-metastatic groups, respectively). Increasing PSA (OR=4.59, p<0.001), ALP (OR=4.24, p<0.001), DRE risk group (OR=1.60, p<0.05), and International Society of Urological Pathology (ISUP) (OR=1.72, p<0.05) were associated with increasing risk of bone metastasis. A two-unit rise along the D-Amico risk strata was associated with a 28-fold increase in the odds of metastasis (low risk vs high risk, OR=29.56, p<0.001; low risk vs intermediate risk, OR=2.80, p=0.368). Among participants with more than 30% of their core-biopsy volume involved with adenocarcinoma, 54.0% had bone metastasis (OR=1.33, p=0.06). Also, 57% of those with perineural invasion had bone metastasis (OR=4.0, p=0.01). Perivascular invasion was not a statistically significant predictor (p=0.346). All patients with cribriform pattern histology had bone metastasis (100%; OR=9.40, p=0.04). Of those with bone pain, 48.4% had bone metastasis (OR=2.25, p=0.002). PSA and ALP exhibited strong independent associations with bone metastasis, with PSA outperforming other predictors (AUC under ROC curve of 81.65% vs 77.97% for ALP, 69.73% for DRE, and 79.19% for ISUP. On multivariate logistic regression, the combined AUC for PSA and ALP was 89.28%, while that for combined PSA, ALP, DRE, and ISUP was 92.3%). A PSA cut-off of 18.95 ng/mL or an ALP cutoff of 59.48 IU/L, individually, detected 97.5% of bone metastasis. Combining a PSA cut-off of 20.85 ng/mL with an ALP cut-off of 44.0 IU/L yielded a 100% detection rate, while PSA above 20 ng/mL, DRE >T2c, and Gleason score >7 predicted 95% of bone metastasis in this cohort. In a logistic regression model, ALP, ISUP, and DRE stage significantly improved the bone metastasis detection rate of PSA in prostate cancer (z-statistic gain at p=0.05 was 2.41; new AUC=92.29%). The inclusion of bone pain, cribriform pattern histology, perineural invasion, and percentage core involvement of adenocarcinoma yielded just marginal gains (maximum z-statistic gain at p=0.05 was 0.61; new AUC: 93.90%). Conclusion For high-risk prostate cancer, bone scans are recommended. In Ghana, a PSA cut-off of 18.95 ng/mL could safely exclude 20% of bone scans, missing only 2.5% of bone metastases, saving $170 per patient ($55,000 nationally per year). Combining this with an ALP cut-off of 44.0 IU/L (0% false-negative rate) detects 100%. A validated risk model combining PSA, ALP, ISUP, and DRE achieves an AUC of 92.3%, sensitivity of 89.9%, specificity of 78.8%, and accuracy of 85.6%, offering a practical, digitally deployed, decision support tool.