This was a retrospective cohort study of Surveillance, Epidemiology, and End Results (SEER)-Medicare patients diagnosed with mRCC and matched non-cancer controls from 2007–2015 (Figure 1). The SEER-Medicare database is composed of patients with cancer identified in the SEER cancer registry who have Medicare insurance coverage. Non-cancer controls were identified from the ‘non-cancer’ Medicare 5% sample which is comprised of a random 5% sample of patients with Medicare and excludes all patients in the SEER cancer registry. Patients with mRCC were required to have had at least twelve months of continuous enrollment in Medicare fee-for-service parts A (inpatient facility, skilled nursing facility, hospice and home health services) and B (medically necessary doctor’s services, outpatient care, and some preventative care) before their mRCC diagnosis, and in Parts A, B, and D (prescription medication coverage) for twelve months after the metastatic index date or until death. Patients with mRCC were excluded if their metastatic diagnosis occurred at autopsy or death, if they were less than 65 years of age at metastatic index date, or if they had a second primary diagnosis of cancer at another site between the initial SEER RCC diagnosis date (if they were initially diagnosed at Stage I, II, or III) and their metastatic diagnosis date. Patients were excluded from analysis if they had a claim with a diagnosis code indicating MCI/D (Appendix A) at any time during the twelve months prior to their date of metastatic diagnosis. Non-cancer matched controls selected from the Medicare 5% sample were required to be at least 65 years of age on July 1 of the year of their matched case’s metastatic diagnosis, which served as the control’s cohort entry date. Controls were required to be enrolled in Medicare fee-for-service parts A and B in the twelve months prior to July 1 of their entry year, and for twelve months following their entry or until death. Potential controls were excluded if they had a claim with a diagnosis code in any position indicating MCI/D (Appendix A) in the twelve months prior to cohort entry. Three controls were selected per case, and controls were matched using greedy match without replacement on year of entry (+/− two years), patient age (+/− five years), patient sex, patient race (matched Black/not Black), hypertension, cerebrovascular disease, hemiplegia/paraplegia and Charlson comorbidity score (+/− two) calculated from the twelve months preceding their entry date.

Patients were determined to have an incident diagnosis of mild cognitive impairment or dementia if they had a diagnosis code for MCI/D in any diagnosis position on an inpatient, outpatient, carrier, or home health Medicare claim in the twelve months following the patient’s cohort entry date (Appendix A).³⁸ This cohort likely includes patients with cancer-related dementia and mild cognitive impairment as well as patients with MCI/D that may be unrelated to cancer. The date of the earliest claim with one of these codes served as the patient’s MCI/D diagnosis date.

Patient demographics at metastatic diagnosis or cohort entry included self-reported race/ethnicity, age, sex, geographic region of residence, residence in a metropolitan region as defined by Rural-Urban Continuum Codes, and ZIP code of residence-level socioeconomic characteristics. ZIP-code-level socioeconomic characteristics are continuous variables that are presented in this analysis as quartiles, with quartile 4 (Q4) representing areas with the highest rates of households experiencing poverty, the highest percentage of Black residents, and the highest percentage of adults without a high school education. Validated coding algorithms were used to assess patient comorbidities of interest in the twelve months prior to the metastatic index date or cohort entry date using diagnosis codes (Appendix B) from inpatient, outpatient, and carrier Medicare claims files.³⁹ For patients with mRCC, we additionally assessed stage at initial RCC diagnosis, histology at initial RCC diagnosis, and marital status at initial RCC diagnosis.

Utilization of an OAA was determined by the presence of a Part D prescription drug fill record for any of the following OAAs in the twelve months following metastatic diagnosis: sorafenib, sunitinib, pazopanib, everolimus, and axitinib. OAA utilization was treated as a binary time-varying exposure (never used/ever used an OAA), with the date of a patient’s first OAA prescription claim serving as the date at which they began using an OAA.

Receipt of a partial or radical nephrectomy was assessed among RCC patients in the twelve months prior to and twelve months following metastatic diagnosis. Nephrectomies were identified by the presence of one of the following Healthcare Common Procedure Classification System (HCPCS) codes indicating nephrectomy on a Medicare carrier or outpatient claim in that time period: 50230, 50545, 50546, 50220, 50225, 50240, 50280, 50290, 50542, or 50543. Receipt of nephrectomy was coded as a time-varying exposure (has not had nephrectomy/had nephrectomy) beginning at the patient’s metastatic diagnosis date. The date of the first Medicare claim with a nephrectomy HCPCS code served as the date at which a patient was classified as having a nephrectomy.

Patient demographics and clinical characteristics were presented stratified by case status. Continuous variables were described using mean (SD) or median (Q1, Q3). Categorical variables were described using N (%) and tested for differences between matched cases and controls using chi-square tests, Wilcoxon rank-sum tests/t-tests and standardized differences to assess the matches. Patient ZIP code, local education percentages, poverty percentages, and racial composition percentages were analyzed as quartiles. Baseline differences between the mRCC and non-cancer cohorts were assessed with standardized difference scores. The threshold for reporting was set at 10%, given that a standardized difference of less than 10% suggests balance between the case and control groups with respect to the variable in question.⁴⁰

Cumulative incidence of MCI/D was calculated for mRCC patients and controls at one year from the cohort entry date based on estimates from the cumulative incidence function, to account for the high risk of death in this patient population. Patients were censored at one year following their cohort entry or at their death, whichever occurred first. The one-year cumulative incidence was stratified by case status, age group, and patient race/ethnicity. Group differences in the cumulative incidence function were evaluated using Gray’s test. Kaplan-Meier curves were used to test the proportional hazards assumption.

Unadjusted and multivariable-adjusted Cox proportional hazards regression analyses were used to estimate the associations between mRCC case status and risk of developing MCI/D. Adjustment variables were pre-selected based on prior literature and clinical input. Models were adjusted for the following characteristics at the metastatic index date (mRCC cohort) or cohort entry (controls): patient age in years, sex, race/ethnicity, year of cohort entry, residence in a metropolitan area, geographic region, dual-enrollment in Medicaid and Medicare, previously-defined ZIP code-level socioeconomic measures, and patient comorbid conditions. In a post-hoc analysis to assess for potential residual confounding in the association between mRCC and MCI/D development, we tested the association between mRCC and a falsification endpoint, hospitalization for congestive heart failure (CHF) (Appendix A).

A cohort limited to mRCC patients was used to assess whether RCC-specific therapies increased the risk of developing MCI/D among mRCC patients in the twelve months following metastatic diagnosis. Univariable and multivariable adjusted time-varying Cox proportional hazards regression analyses were used to assess the associations between development of MCI/D and receipt of OAAs and receipt of partial or radical nephrectomy (as time-varying exposures), with additional adjustment for patient age, race/ethnicity, and other patient demographic and clinical factors, as previously described.

An alpha level of 0.05 was used to establish statistical significance of tests, and all tests were two-sided. All analyses were conducted using SAS version 9.4 (Cary, NC). The study protocol was reviewed by the Duke University Health System Institutional Review Board. It was initially approved and later determined to be exempt (Protocol Pro00101962).

There were 2,792 patients who received an mRCC diagnosis in the given time period and who were over the age of 65 with the required periods of continuous enrollment both before and after diagnosis identified from the SEER-Medicare database. Of those patients, 259 with a prevalent diagnosis of MCI/D prior to their metastatic diagnosis were excluded, leaving 2,533 patients with mRCC. The final cohort consisted of 2,533 patients with mRCC and 7,027 non-cancer controls (Figure 1).

Demographic variables were compared between patients with mRCC and non-cancer controls. Cases and controls were generally well-matched based on directly matched factors. Among all analyzed beneficiaries, mean age was 76.2 years (standard deviation = 6.7 years) and 42.3% were female. Patients with mRCC were less likely than controls to live in ZIP codes with adults who had less than a high school education (standardized difference 17.7%) and where households had incomes below the poverty line (standardized difference 16.0%). Patients with mRCC had higher incidence of mild liver disease (standardized difference 31.2%) and renal disease (standardized difference 20.0%) (Table 1).

Median available follow-up time in the cohort was 365 days (IQR 341 days– 365 days). The one-year cumulative incidence of MCI/D was greater in patients with mRCC than in non-cancer controls (6.0% vs 1.3%, Gray’s P<.001) (Figure 2a). When patients with mRCC and patients without cancer were analyzed as one cohort, there was greater cumulative incidence of MCI/D in older patients (Gray’s P<.001) (Figure 2b). There was not a statistically significant difference in cumulative incidence of MCI/D based on race/ethnicity (Gray’s P=0.067) (Figure 2c).

Fully-adjusted Cox proportional hazards regression was used to identify predictors of incident MCI/D in the twelve months following mRCC diagnosis or entry to the case-matched cohort. This model included both patients with a diagnosis of mRCC and the non-cancer controls. Diagnosis of mRCC was a significant positive predictor of incident MCI/D (hazard ratio [HR] 8.52, 95% confidence interval [CI] 6.49–11.19, p<.001) (Table 2). Increased age (HR 1.05 per one-year, 95% CI 1.03–1.07, p<.001) as well as later year of cohort entry (HR 1.21 per one-year, 95% CI 1.14–1.28, p<.001) were significant positive predictors of incident MCI/D. Multiple comorbidities were associated with a higher hazard of incident MCI/D, including CHF (HR 1.44, 95% CI 1.07–1.94, p=.016), cerebrovascular disease (HR 1.40, 95% CI 1.05–1.87, p=.022), rheumatologic disease (HR 1.63, 95% CI 1.06–2.49, p=.025) and hemiplegia/paraplegia (HR 2.04, 95% CI 1.03–4.04, p=.042). Mild liver disease was found to be a negative predictor of MCI/D (HR 0.59, 95% CI 0.37–0.94, p=.027).

In the post-hoc falsification endpoint analysis, there was a borderline significant association between mRCC and CHF hospitalization (HR 1.32, 95% CI 1.01–1.72, p=.045).

A second fully-adjusted Cox proportional hazards regression was used to identify predictors of incident MCI in the twelve months following mRCC diagnosis. This model included only those patients with a diagnosis of mRCC. Neither use of OAAs (HR 0.89, 95% CI 0.59–1.34) nor undergoing partial or radical nephrectomy (HR 0.78, 95% CI 0.52–1.17) was a significant predictor of incident MCI/D (Table 3). Increased age was a significant predictor of incident MCI/D (HR 1.06 per one-year, 95% CI 1.03–1.08, p<.001), as was self-identifying as non-Hispanic and Black when compared to those identifying as non-Hispanic and White (HR 1.92, 95% CI 1.02–3.59, p=.042). Mild liver disease was found to be a negative predictor of incident MCI/D (HR 0.55, 95% CI 0.32–0.95, p=.033).

Many of this study’s limitations are inherent to the SEER-Medicare database. First, we were unable assess renal function in our mRCC or non-cancer control cohorts because the database is not linked to patient laboratory data. Renal dysfunction may contribute to the increased incidence of MCI/D seen in patients with mRCC; future studies might include serum calcium or creatinine levels, or creatinine clearance in the regression model used to predict incident MCI/D. Second, it was challenging to evaluate the impact of prior cancer treatments on MCI/D and to identify patients with current cancer diagnoses given that our database was limited to a one-year look-back period. Next, the ability to identify the progression of localized disease to metastatic disease using SEER-Medicare data is not complete.⁵⁷ Thus, there may be patients who were initially diagnosed with stage I, II or III RCC and who progressed to mRCC who were not included in this study. Finally, we cannot account for the impact of individuals’ education on cognition because the SEER-Medicare database does not record this information.

There are also limitations inherent to claims analysis. Some patients with mRCC who experienced cognitive deficits may not have been assigned MCI/D International Classification of Diseases (ICD) diagnosis codes in the SEER-Medicare database because of coding practices in a real-world sample. These deficits may have been identified objectively on neuropsychological tests or subjectively by patients themselves, and can substantially impact the quality of life experienced by these patients.³³ However, given that both patients with mRCC and non-cancer controls were subject to this limitation, it is unlikely that the inability to capture all MCI/D cases alters the difference in incident MCI/D that we detected between groups. Though we are unaware of existing studies that validate the ICD codes for mild cognitive impairment, this does not preclude impactful research findings utilizing these claims.

We acknowledge that a portion of the increased incidence in MCI/D observed in patients with mRCC may be the result of residual confounding or detection bias. While we did observe a borderline significant association between mRCC and our CHF hospitalization endpoint, the magnitude of the hazard ratio and strength of the association seen in the p-value were both considerably smaller than those observed for the association between mRCC and incident MCI/D. Mild cognitive impairment or dementia may have been more frequently diagnosed in patients with mRCC in the setting of more frequent contact with the healthcare system. If so, this bias would overestimate the hazard ratio for developing MCI/D in those patients with mRCC. The literature describes the potential for this bias, but confirms these measures still provide valuable insight into patients suffering from MCI/D.^58,59

Finally, the control group selected in this study comprises patients who did not have any diagnosis codes related to cancer. This is similar to previous works that evaluated the effect of cancer on cognitive decline.^60–62 While this study does not offer insight into the possible differences in MCI/D seen in patients with different types of cancer, its relevance lies in establishing the evidence for providers to have a lower threshold for cognitive screening in patients who are newly diagnosed with mRCC.

Older adults with mRCC have higher one-year cumulative incidence of MCI/D than those without cancer. This finding is significant to primary care providers, oncologists and surgeons in that each must be aware of the increased risk for cognitive decline in their patients with mRCC. It may be beneficial to integrate cognitive screening in treatment models. In addition, involvement of patients’ families shortly after diagnosis can be encouraged in the event that patients become unable to make decisions for themselves.

Among patients with mRCC, neither treatment with OAAs nor partial or radical nephrectomy was associated with an increased risk of developing MCI/D. These findings suggest that while a diagnosis of mRCC in older adults should raise concern for the possibility of developing MCI/D, neither of these treatment options should be avoided in the hope of avoiding MCI/D.