Data source
Data were accessed from SA-PCCOC registry [14]. SA-PCCOC is a disease-specific, multi-institutional, prospective registry which has been operating since 1998. Since 2008, the registry has been expanded to collect clinical and oncologic data from both public and private treatment centres and captures data from approximately 90% of men with prostate cancer in South Australia. While intervals for collecting PROMs have changed over time, baseline and 12 months PROMs have been consistently collected since the registry’s inception.
Sampling
This study included men enrolled in SA-PCCOC who were diagnosed between 2010 and 2019 and had completed any items in the PROMS surveys at baseline or 12 months post treatment. This period was selected to (1) reflect contemporary treatments practices, (2) approximate complete population-based coverage, and (3) be inclusive of both public and private patients. In total, 8143 men were enrolled in SA-PCCOC from 2010 to 2019, of whom 6030 had participated baseline and/or 12-month PROMs surveys. Those who received RP, EBRT, brachytherapy or active surveillance were included in analyses (n = 4926). Men undergoing primary hormonal treatment (n = 256), chemotherapy (n = 8), other/unknown treatment (n = 724) or with missing treatment data (n = 116) were excluded (Fig. 1).
Measurement and variables
Outcomes
The first set of outcome variables were the functional outcome scores: at baseline (pre-treatment) and 12 months after treatment commenced. Functional outcomes were measured using Expanded Prostate Cancer Index Composite (EPIC)-26 domain scores [15]. Four items were used in the urinary incontinence domain, four items for urinary irritative/obstructive symptoms, six items for bowel function, and six items for sexual function. Each functional domain was calculated in accordance with previous EPIC-26 guidelines [16]. Accordingly, if \(\ge\)20% of the items in a domain were missing a response, that domain was not calculated. Domain scores range from 0 to 100 with a higher score indicating better function. An established convention of minimal clinically important difference (MCID) in EPIC-26 domain score of PROMs [17] was used to describe and complement the reported scores. MCIDs were set at 10-point score differences for sexual function, 6-point for urinary incontinence, 5-point for urinary obstruction, and 4-point for bowel function.
The second set of outcome variables were extent of “bother” regarding urinary, bowel and sexual function. Each bother outcome was measured using a single item question from the EPIC-26 that asked men how big a problem their [urinary symptoms, bowel habits, lack of sexual function] had been for them in the last four weeks. The responses were: no signs, very small, small, moderate, or big bothers. For descriptive reporting, we merged “no signs”, “very small” and “small” together and “moderate” with “big” bother but retained the original five responses in the final multivariable analyses.
Treatments
Primary treatments were grouped as RP, EBRT, brachytherapy and active surveillance based on predefined criteria used by SA-PCCOC. Men were categorised as having active surveillance if this was recorded as their initial treatment in their medical record or derived through registry algorithms based on risk level and clinical characteristics. Between 2010 and 2016, active surveillance protocol would include a repeat ultrasound guided transrectal or transperineal biopsy by 12 months, with no routine use of magnetic resonance imaging (MRI) for surveillance. From 2017, MRI+/- transperineal biopsy became the more common practice. Adjuvant and salvage therapies were not accounted for due to inconsistent collection of secondary treatment data in SA-PCCOC registry.
Covariates
Sociodemographic characteristics included age at diagnosis (below 60, 60–64, 65–69, above 70) and socioeconomic status based on the 2016 Socio-Economic Indexes for Areas (SEIFA) score. SEIFA is an indicator of relative economic and social advantage/disadvantage of areas based on a range of socio-economic factors such as employment status, income and household status [18]. SEIFA scores were grouped into quintiles (lowest, low, medium, high, and highest socio-economic status). Risk classification was according to 2017 National Comprehensive Cancer Network (NCCN) groupings [19] and was derived from prostate-specific antigen (PSA) level, Gleason grade group and clinical stage of the disease; high risk (clinical stage T3a OR PSA > 20 ng/mL OR Gleason score of 8 or more), intermediate risk (clinical stage T2b and T2c OR PSA 10–20 ng/mL OR Gleason score of 7 (3 + 4 or 4 + 3)), and low risk (clinical stage T2a or earlier AND PSA < 10 ng/mL AND a Gleason score of less than 7). Diagnostic PSA (< 4, 4–10, 10–20 and > 20 ng/mL) and total Gleason score ( < = 6, 3 + 4, 4 + 3, 8 and 9–10) were coded as previously reported [11].
Statistical analysis
Patient characteristics were summarized using descriptive statistics (frequency counts with percentages for categorical variables, means and standard deviations for normally distributed continuous variables, and median and interquartile range for continuous variables that were not normally distributed).
Functional outcomes were missing in a high proportion of men enrolled in SA-PCCOC, due to some surveys not being sent within specified time periods, survey non-response, or variability in response rates to specific EPIC-26 items. To derive population-wide estimates of functional scores we applied inverse probability weighting to all analyses. Weights were calculated by deriving propensity scores for ‘being a respondent’ for each specific functional outcome, using logistic regression to predict likelihood (propensity) of being a respondent within the entire SA-PCCOC cohort based on patient characteristics (with weight = 1/propensity score). The use of inverse probability weighting to account for missingness has been suggested in previous literature [20].
Multivariable linear regression models were fitted to estimate the effect of EBRT, brachytherapy, and active surveillance vs. RP on the 12 months functional outcome scores. Multivariate ordinal logistic regression models were fitted to identify the association between treatment types and patient-reported bother at 12 months. RP, which is the largest category, was considered as a reference group. Additional models were run with active surveillance as the reference group. Separate analyses were undertaken for each of the functional outcomes and bother items, with each model including only those participants with complete data for that specific outcome (number provided in Fig. 1). All the models were adjusted for covariates (age, SEIFA score, PSA level, risk category and primary symptoms) and respective baseline functional measures. Collinearity was checked using variance inflation factor (VIF) leading to “total Gleason score” being removed from the final adjusted models in favour of “risk category”. The final models present the mean difference in functional scores at 12 months between treatments after adjustment for baseline scores and other covariates, and adjusted odds ratio (aOR) for greater bother at 12 months, along with confidences intervals (CI) and p-values. Adjusted mean values of the outcome variables in each treatment category were calculated from postestimation predictions following multivariable regression models. Statistical significance was regarded as two-tailed \(p<0.05\). All analyses were carried out using Stata version 15.0 (StataCorp, College Station, Tx, USA).