• 2019-07
  • 2019-08
  • 2019-09
  • 2019-10
  • 2019-11
  • 2020-03
  • 2020-07
  • 2020-08
  • 2021-03
  • br Exposure Treatment Choice br The


    Exposure: Treatment Choice
    The primary exposure was the patient’s primary treatment (ie, conservative management, surgery, or radiation). Conservative management included patients on active surveillance, watchful waiting, and primary androgen deprivation therapy since SEER does not differentiate between these approaches. For conserva-tive management patients, the “treatment” date was considered their date of diagnosis. Patients who received both surgery and radiation were assigned to their initial treatment.
    We categorized 3X FLAG as white and non-white due to limited num-bers of non-whites (ie, African American, Asian, Hispanic, North American Native, and other). We report total number of comorbidities, but matched on specific comorbidities in the pro-pensity score analyses. The MHOS does not include a direct measure of cognitive function. If the person completing the sur-vey indicated they were serving as a proxy for the sampled per-son, we used that as a marker of poor cognitive function.
    Statistical Analysis
    We first compared patient and regional characteristics among non-cancer patients and each treatment separately using chi-square tests. We then examined patient functional status among noncancer patients and each treatment using chi-square tests for categorical variables and Wilcoxon rank sum tests for continuous variables. For statistical modeling, we had to address 2 issues: (1) missing values in covariates, including marital status, household income, and education; and (2) covariate imbalance across cancer and noncancer patients. To address the first issue, we used a multiple imputation approach and created five imputed datasets for noncancer patients and for each treatment group. Imputations were done separately for noncancer and cancer patients. To address the second issue, we sepa-rately used 1:5 propensity score matching. For each treatment group (ie, conservative management, surgery, and radiation), we fit a sepa-rate propensity score model to estimate the propensity of having can-cer. To do this, we fit a logistic regression model with cancer/ noncancer as the binary outcome and survey year, survey month, age, race, marital status, household income, education, each specific comorbidity, region, and proxy as covariates to estimate the propen-sity of having cancer. Once the propensity score was estimated, we used 1:5 matching based on these propensity scores to select the non-cancer patients. We matched by survey year and survey month to ensure that noncancer patients had similar time intervals between their 2 surveys as the treated patients with whom they were matched. To obtain the best results, we matched noncancer patients to each
    treatment separately, which is an important distinction from prior work.13 The noncancer control groups for each treatment type were not mutually exclusive, so noncancer patients could be used in multi-ple comparison groups if they represented the best match.
    We then used separate generalized linear mixed models with change in functional status as the dependent variable and treat-ment and control designation as the independent variable, adjusting for the clustering due to matched unit (represents the 1 treated cancer patient and the 5 noncancer patients involved in the 1:5 matching) and insurance plan by treating them as ran-dom effects in the model. This accounts for the multilevel struc-ture of the data (patient level, matched units, and insurance plan). This process was repeated for each imputed pair of control and treatment datasets, and results were combined using Rubin’s formula implemented via SAS procedure MIANALYZE.17
    We performed all analyses using SAS v9.4 (Cary, NC). All tests were two-sided, and the probability of a type I error was set at 0.05. The University of Pittsburgh institutional review board exempted this study from full board review.
    We included 40,177 noncancer patients and 408 prostate cancer patients in our study (Table 1). Of the 408 cancer patients, 143 chose conservative management, 59 received surgery, and 206 received radiation. All patients completed surveys assessing func-tional status at 2 different times. The cancer patients each com-pleted 1 survey prior to treatment and 1 survey after treatment. All four groups of patients were similar in terms of number of comor-bidities and education (both P > .05). Patients receiving surgery were generally younger whereas those undergoing conservative management were older compared with noncancer patients (both P < .001). A higher proportion of surgery and radiation patients were married compared with their noncancer peers (both P < .05). The vast majority of patients across all groups completed their own surveys. Surgery patients had higher baseline physical function as measured by ADL scores (mean § standard deviation of 4 § 12 compared with 7 § 15, 9 § 16, and 6 § 13 for noncancer, conser-vative management, and radiation patients, respectively; P = .01) and PCS scores (mean § standard deviation of 48 § 9 compared with 43 § 11, 43 § 11, and 44 § 11 for noncancer, conservative management, and radiation patients, respectively; P < .001).