Introduction and objective
It is unclear how the interaction of clinical covariates modulates medical treatment response in individual profiles of men with lower urinary tract symptoms/benign prostatic enlargement (LUTS/BPE). We developed a predictive analytics solution using baseline characteristics that commonly define patient risk of disease progression to project the change in storage, voiding and nocturia LUTS, in individual profiles receiving placebo (PBO), dutasteride (DUT), tamsulosin (TAM) or DUT/TAM combination therapy (CT).
Methods
A total of 9167 men with LUTS/BPE at risk of progression from three PBO-controlled DUT trials and one comparing DUT, TAM, and CT were included in the analyses to predict response to placebo up to 24 mo and active treatment up to 48 mo. Baseline characteristics included as predictors were: age, International Prostate Symptom Score (IPSS), total prostate volume (PV), maximum urine flow rate (Qmax), prostate-specific antigen (PSA), postvoid residual urine (PVR), α-blocker (AB) usage within 12 mo, and randomized treatment. A generalized least-squares model was developed for longitudinal IPSS storage and voiding sub-scores. For nocturia IPSS-Q7 a Generalized Additive Mixed Model was used to generate predictions. A ‘placebo response’ model was added to benchmark the predicted outcomes to treatment for 2-year f/u.
Results
For most profiles, nocturia, voiding and storage symptoms improved significantly with CT compared to DUT or TAM therapies. Higher PSA and PV values predicted a greater improvement in the outcomes explored in DUT compared to PBO. Higher PSA levels predicted a greater improvement in nocturia with all active treatments. Moreover, higher Qmax levels predicted better nocturia and voiding scores with all active treatments. Higher PVR levels were predictive of lower improvements in voiding scores with all active treatments. Likewise, previous use of AB indicates a lesser improvement in nocturia with TAM compared to CT therapy and on the storage scores with all active treatments. Also, higher PV levels had worse voiding scores with TAM in comparison with CT and DUT. The models were deployed in an interactive web app to help visualize and understand results for any possible combination of predictors up to 4 years.
Conclusions
This predictive modelling based on large data sets contributes to understand how risk factors for disease progression interact and affect treatment impact on different type of symptoms, reinforcing the importance of an individualized approach for LUTS/BPE management.
Source of funding
GSK (study 219073)
