This multicenter retrospective cohort study will use routinely collected clinical data from the PLA General Hospital consortium (the First, Third, Fourth, Fifth, Sixth, Seventh, and Eighth Medical Centers) spanning 2001 to 2021. Patients will be divided into 2 cohorts: cohort 1 (First Medical Center) and cohort 2 (the remaining medical centers). Eligible patients include those with a confirmed diagnosis of BPH and long-term follow-up, with the date of first diagnosis serving as the baseline. Patients with prostate cancer, urethral stricture, neurogenic bladder, or missing key clinical data will be excluded.

Routinely collected data will be extracted from each center’s electronic health record (EHR) system, including outpatient, inpatient, and health examination records. Diagnoses and end point events will be identified using prespecified International Classification of Diseases, 9th and 10th Revision, codes and procedure code lists (Table 1), and relevant laboratory and clinical measurements will be retrieved from structured records. Deidentified datasets will then be transferred to the coordinating center for harmonization before analysis (Figure 1).

To evaluate model transportability beyond hospital-based EHRs, we will use the China Health and Retirement Longitudinal Study (CHARLS), a nationally representative longitudinal cohort, as an independent external validation dataset. Variables and end points will be harmonized between the PLA General Hospital EHR cohort and the CHARLS using prespecified definitions; external validation will be conducted for predictors and end points that can be consistently operationalized in the CHARLS.

The outcomes of this study are (1) BPH-related surgery (transurethral resection of the prostate and bipolar or laser procedures), (2) urinary retention, (3) inguinal hernia, (4) chronic kidney disease (CKD; estimated glomerular filtration rate <60 mL per minute per 1.73 m² persisting for ≥90 days), (5) urothelial carcinoma, and (6) all-cause mortality (Table 1).

The inclusion and exclusion criteria are outlined in Textbox 1.

Baseline variables used for outcome analyses and risk stratification modeling will include residence, age (years), BMI (kg/m²), smoking status (yes or no), alcohol consumption (yes or no), comorbidities (hypertension, diabetes mellitus, coronary artery disease, psychiatric disorders, sleep disorders, respiratory diseases, and constipation), and laboratory or clinical measures (serum creatinine, serum uric acid, triglyceride-to–high-density lipoprotein cholesterol ratio, neutrophil-to-lymphocyte ratio, alanine aminotransferase, D-dimer, total prostate-specific antigen, free prostate-specific antigen, and prostate volume).

Rigorous measures will be implemented to guarantee data quality, integrity, confidentiality, and statistical validity:

On the basis of Cox regression and Fine-Gray competing risk models, at least 300 to 490 events are required to analyze approximately 25 covariates [38-40]. Assuming a 10% incidence of surgery at 5 years, approximately 3334 patients will be needed. For rarer outcomes such as CKD (3% incidence), approximately 10,000 patients are required. The final sample size will be refined according to empirical event rates [40].

If, after data cleaning and cohort assembly, the effective sample size or event numbers for any prespecified end point are insufficient, we will implement the following prespecified strategies: (1) prioritization of end points with adequate events for multivariable modeling while reporting end points with sparse events descriptively and/or as exploratory analyses; (2) simplification of model specification by restricting covariates to a clinically justified core set, combining sparse categories when appropriate and using shrinkage or penalized methods to improve stability and minimize overfitting; and (3) use of alternative time-to-event approaches where needed, including penalized or stratified Cox models or time-varying effects for Cox analyses and cause-specific hazard models (or cumulative incidence estimation without multivariable regression) as alternatives when Fine-Gray models are unstable due to sparse subdistribution events. Any deviations from the initial plan and their rationale will be transparently reported.

To address the study objectives, analyses will comprise (1) long-term outcome profiling, (2) association analyses, and (3) prognostic risk modeling. Time-to-event methods will be used to summarize incidence, cumulative incidence, and median time to event, as appropriate. For association analyses, Cox proportional hazard regression will be applied for all-cause mortality, whereas Fine-Gray competing risk models will be used for nonmortality end points where death may act as a competing event, reporting hazard ratios or subdistribution hazard ratios with 95% CIs. Baseline variables for analysis and modeling are specified in the Baseline Variables for Analysis section. For prognostic risk modeling, multivariable prediction models will be developed and validated to support risk stratification, with performance assessed using discrimination (eg, concordance index or area under the curve), calibration, and clinical utility metrics (eg, decision curve analysis), as applicable.

Cox proportional hazard regression is appropriate for this study because the primary end points are time-to-event outcomes with right censoring and the Cox model provides an efficient and interpretable framework to account for censoring while modeling risk over follow-up. In this protocol, Cox regression is used within a prognostic (risk stratification) framework rather than solely for etiologic inference. Cox models will be applied to all-cause mortality (event 6). For nonmortality end points where death may act as a competing event, Fine-Gray competing risk models will be used to support risk estimation based on cumulative incidence.

If supported by the empirical data structure and transition processes (eg, sufficient event counts and identifiable transition pathways), we will further explore and construct multistate models to characterize dynamic transitions between clinical states (Figure 2).

Standardized case report forms will be extracted by trained staff at each center using a unified procedure and securely transferred to the coordinating center via encrypted channels. The coordinating center will perform logic checks and randomly audit 10% of records to monitor data consistency (target inconsistency rate of ≤3%). Outliers will be identified during data cleaning and handled according to prespecified rules. Missing data will be handled using multiple imputation for all variables with missing values, and an overall sensitivity analysis will be conducted using conditional imputation. Preliminary checks indicate that most variables have <20% missingness (many with none), and all variables have <30% missingness.

Baseline characteristics will be summarized as means and SDs, medians and IQRs, or counts and percentages (Table 2).

Time-to-event distributions will be summarized using appropriate survival and competing risk methods, with effect estimates from univariable and multivariable models reported as hazard ratios or subdistribution hazard ratios with 95% CIs (Multimedia Appendix 1).

Prediction models (eg, nomograms) will be constructed from multivariable results. Internal validation will use bootstrapping. External validation will include internal-external cross-validation between the 2 cohorts and an independent transportability assessment on the CHARLS (for predictors and end points that can be consistently harmonized). If miscalibration is observed in the CHARLS, we will report recalibration approaches (eg, intercept or slope updating) and present both original and recalibrated performance. Model performance will be assessed using concordance index, receiver operating characteristic curves, area under the curve, calibration plots, decision curves, and reclassification metrics (net reclassification improvement and integrated discrimination improvement).

Statistical significance will be defined as 2-sided tests, with P<.10 (univariate analysis) and P<.05 (multivariate analysis) considered significant.

As this study is a retrospective cohort analysis based on routinely collected follow-up data, patients and the public were not directly involved in formulating the research question, designing the study, or conducting the analyses. Nevertheless, we place strong emphasis on the clinical relevance and accessibility of the findings. Upon completion, results will be disseminated through peer-reviewed publications and scientific conferences and will also be shared with urology-related patient organizations and professional communities in China to facilitate translation into routine clinical decision-making and better reflect patient priorities. Building on these findings, we plan to explore prospective studies that incorporate patient and public involvement to further strengthen clinical relevance and societal impact.

The study protocol has been approved by the ethics committee of the PLA General Hospital (approval S2022-700-01). The entire study will be conducted in accordance with the principles of the Declaration of Helsinki and relevant ethical guidelines for medical research.

As this study is a retrospective secondary analysis of routinely collected electronic health record data, the Ethics Committee of the PLA General Hospital granted a waiver of informed consent because the analysis uses deidentified data, involves minimal risk to participants, and no direct contact with participants will occur. No participant compensation will be provided.

The study will use routinely collected clinical data for research purposes only. Access to the dataset will be restricted to authorized team members, thereby minimizing the risk of privacy breaches. Data protection measures (including encryption, deidentification, and secure storage and transfer) are described elsewhere in the Methods section.

Findings will be disseminated through publications in peer-reviewed international and domestic journals and presented at scientific conferences, targeting clinicians, researchers, and the broader public to share key insights into the long-term prognosis of patients with BPH. Authorship will comply with the recommendations of the International Committee of Medical Journal Editors to ensure academic integrity and recognition of contributions.

Upon study completion, deidentified follow-up datasets will be archived in the Clinical Research Data Platform of the PLA General Hospital. Internal team members may conduct further subgroup analyses. External researchers may apply for access by submitting a written request detailing study aims and protocol. Access will be granted only after review and approval by the ethics committee and data platform and upon signing a data use agreement specifying scope and confidentiality obligations. Data will be retained for 20 years after study completion, after which they will be permanently destroyed according to institutional policy.

Research on the clinical course and outcomes of BPH has several limitations. First, although there is existing evidence suggesting that frailty phenotype and larger prostate volume are established risk factors for disease progression [41,42], robust quantitative data on mid- to long-term risks of key outcomes, such as surgical intervention and acute urinary retention, remain limited. Second, most studies have relatively short follow-up durations, lacking long-term risk estimates over 10 years, particularly for older adult and frail populations [43-45]. Third, many prior studies have small sample sizes or single-center designs, limiting statistical power for infrequent outcomes such as CKD [46,47]. More importantly, inadequate adjustment for competing risks in some analyses may lead to overestimation of disease-specific outcomes [48,49]. These limitations highlight the urgent need for large-sample, long-term, multicenter studies using robust statistical frameworks.

Compared with previous studies, this work offers several methodological advantages. The follow-up duration is extended, and leveraging nearly 2 decades of clinical data, this study can delineate long-term trajectories of BPH outcomes, overcoming the constraints of short- to midterm studies. The study will also use a multicenter, large-scale cohort of patients across various disease stages and treatment pathways, enhancing representativeness and external validity, with a sample size estimation that ensures sufficient power to detect low-incidence outcomes. The outcome assessment is comprehensive—beyond conventional end points such as surgery and urinary retention, the study incorporates inguinal hernia, CKD, urothelial carcinoma, and all-cause mortality, providing a more holistic evaluation of disease burden, thereby yielding further insights into the pathogenesis and progression of BPH and enhancing the robustness of statistical evidence. The statistical methodology is rigorous—depending on outcome characteristics, both Cox proportional hazard and Fine-Gray competing risk models will be used, with multivariable adjustments to minimize confounding and maximize robustness. Finally, the study will use a predictive modeling framework, building on identified risk factors to construct visual predictive tools (eg, nomograms) with internal validation and cross-validation across cohorts and quantify predictive performance using C-statistics. This addresses the lack of reliable prognostic tools for BPH outcomes.

This study is expected to generate several important insights. First, it may identify novel risk factors beyond age and prostate volume, including metabolic parameters, comorbidity profiles, and initial treatment strategies, thereby improving early risk stratification. Second, it will provide authoritative, large-scale estimates of cumulative incidence and median time to event for major outcomes in the Chinese population, offering a robust evidence base for prognosis and health care planning.

While retrospective cohort studies inherently face challenges [50-54]—such as potential data incompleteness from historical records and inherent biases—these limitations point to clear directions for future advancement. Looking ahead, 2 key priorities will drive follow-up efforts.

The first is expanding and enhancing data quality. Future work will integrate multisource data (including community-based patient records, primary care data, and long-term follow-up supplements) to fill gaps in existing datasets, paired with advanced data cleaning techniques to reduce measurement bias. This expansion will improve data accuracy, completeness, and external validity, making findings more applicable to diverse populations beyond hospital-based cohorts.

The second priority is transitioning to prospective research. Building on this study’s theoretical framework and preliminary insights, a dedicated prospective cohort will be designed to collect real-time clinical data—capturing dynamic BPH symptom changes, treatment responses, and long-term outcomes. Notably, leveraging long-term follow-up, multicenter design, and rigorous methodology, this prospective effort is expected to achieve significant breakthroughs in the study of BPH’s natural history and prognosis. Its results are poised to provide high-quality evidence-based support for optimizing patient management, implementing precision interventions, and formulating health care policies.