Study design and organisation

The ORCHiD study design incorporates two distinct approaches: (1) a retrospective cohort design of black, Hispanic and white patients with OC ages 65–75 years (using secondary data from the Surveillance Epidemiology and Ends Results (SEER)-Medicare linked dataset) and (2) a population-based prospective cohort design with primary survey data of black, Hispanic and white patients with OC ages 20–75 years identified from cancer registries in California, Georgia, Kentucky, New York, North Carolina, Texas, Maryland, Florida and Mississippi. In the SEER-Medicare cohort, we will evaluate the association of three HCA dimensions measurable with administrative data (Affordability, Availability and Accessibility) in relation to guideline adherent initial treatment and supportive care and survival. In the prospective cohort, we will evaluate all five HCA dimensions using a combination of administrative claims data and self-reported measures in relation to guideline adherent treatment and survival, as well as patient-reported outcomes (PROs) such as pain, fatigue and depression.

Retrospective cohort study design

We obtained SEER-Medicare data on 9744 black, Hispanic and white patients with OC, stages I–IV, ages 65 years or older, diagnosed with a first primary OC between 2008 and 2015. Patients were excluded if their OC was diagnosed at autopsy or death, or if OC was not either their first or second primary cancer diagnosis. Patients were required to have at least 12 months of continuous enrolment in Medicare fee-for-service parts A and B prior to the SEER diagnosis, and to have at least one Medicare inpatient, outpatient or carrier claim with a diagnosis code for OC (International Classification of Disease ICD-9-CM and ICD-10-CM diagnosis codes 183.0 or C569) within 2 months of the SEER diagnosis; the date of the earliest claim will serve as the patient’s exact OC diagnosis date as SEER provides only the month and year of diagnosis. Patients were required to have continuous fee-for-service Medicare enrolment in the 12 months following their diagnosis date, or until death. The SEER dataset is routinely linked with Medicare claims data for patients with cancer ages 65+, and includes all SEER data items (sociodemographics, cancer stage, subtype, date of initial diagnosis, vital status and survival time), Medicare claims data from the Center for Medicare and Medicaid Services (CMS)16 and linked external datasets. The SEER-Medicare cohort is linked with external data from the US Census Bureau American Community Survey (ACS), the Area Healthcare Resource file and the NCI hospital files to obtain data at the patient, neighbourhood, physician and hospital levels. The ACS census tract and ZIP code file was used to link region-specific socioeconomic data to the claims records; the Hospital file was used to determine whether a treatment facility was affiliated with an NCI designated cancer centre, whether a facility was ever non-accredited, total number of beds, number of Medicare/Medicaid certified beds, participation in Medicare/Medicaid, urban vs rural location, or teaching status as a measure of academic standing. SEER-Medicare provides the Hospital file linked to their identifiers. We then evaluated the three HCA dimensions—Availability, Affordability and Accommodation—that can be feasibly measured in SEER-Medicare.

Prospective cohort study design

We will recruit 1641 black, Hispanic and white patients with OC ages 20–75 years from nine population-based cancer registries including California, Georgia, Kentucky, New York, North Carolina, Texas, Maryland, Florida and Mississippi (figure 2). These registries cover a diverse range of patients, including typically difficult-to-reach populations, such as Appalachian whites in Kentucky and rural blacks in the Mississippi Delta. We chose to focus on this age group in the prospective cohort (20–75 years) because the median age at diagnosis for OC is 63 years, and 55% of cases are younger than 65 years at diagnosis, enabling us to examine OC outcomes among patients not captured in SEER-Medicare. Sociodemographic and clinical data from registry records will be used to identify patients eligible for the study based on: black, Hispanic or white race; between the ages of 20 and 75 at initial diagnosis; a resident of the registry geographical region at the time of diagnosis and initial contact; at least 6 months postdiagnosis; first primary clinically confirmed OC. To ensure that we have an adequate number of racially diverse participants, we will include patients diagnosed with any stage and subtype of OC and adjust for these variables in statistical models. We will also attempt to recruit all black patients with OC in each registry and use a stratified random sampling approach to select white patients age matched to black patients on ages 20–64 and 65–75 years. We will attempt to recruit all Hispanic patients in two registries—Texas and California—that have high Hispanic populations.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Prospective primary cohort study schema. UAB, University of Alabama at Birmingham.

Prospective cohort recruitment

For each of the target state registries, registry staff will review patient data/records to confirm study eligibility. For eligible patients, a letter describing the study will be sent to the treating physician (when required by the registry). The letter will ask the physician to contact the registry if there are reasons the subject should not be contacted. If no reasons are reported or there is no response from the physician within 2–3 weeks, the registries will send eligible participants a letter informing them of their eligibility for the research study along with a preaddressed postage paid response card which may be used to select whether or not they want their names to be released to the ORCHiD study team. In some registries, if a response card is not received within 2 weeks, phone calls will be made by registry staff in an attempt to get verbal permission and only those who agree to have their names released will be forwarded to the ORCHiD study team for recruitment. At regular intervals, the ORCHiD research team will receive a list of eligible participants from participating registries. ORCHiD study staff will mail out a study packet containing an introduction letter, the study brochure, any registry-required documentation, the study consent form and a prepaid envelope. Between 2 and 3 weeks after the study packet is mailed out, the study interviewers will begin making phone calls to participants to enrol them into the study, answer any questions, conduct informed consent and complete the survey. Phone contact will follow a schedule of attempts during day, evening, and weekend hours. The ORCHiD study was granted a waiver of documentation of consent, therefore, verbal consent obtained during the phone call and documented by the interviewer on an institutional review board (IRB)-approved phone script enables the survey interview to occur during the same phone call. In addition, while not required, participants will be encouraged to also provide documentation of consent via an electronic consent form on REDCap or by signing and returning the mailed paper consent form. The study survey is expected to last 55–65 minutes and can be completed at one time during the call or rescheduled, if needed. On completion of the survey, the biospecimen substudy component will be introduced to participants and interested participants will be administered a biospecimen consent and then subsequently mailed the biospecimen collection kit. An incentive of a US$25 gift card will be provided for completing the study survey, and an additional US$20 gift card for returning the biospecimen kit.

Study measures and data collection

To create the unique richness and depth of data for ORCHiD, we will use a variety of data sources as described below to characterise measures at the patient, neighbourhood, physician and hospital levels.

Measuring HCA dimensions

The Penchansky and Thomas framework6 7 provides five dimensions of HCA. As OC outcomes are particularly impacted by access to care, we will use this framework as a comprehensive measure of access using variables relevant to OC that can be measured at either the patient, neighbourhood, provider or health system levels using data sources as outlined in table 1. As part of formative work for ORCHiD, we conducted a comprehensive review of the literature to identify existing instruments to characterise HCA across all five dimensions. However, we noted significant challenges: (1) none of the existing instruments comprehensively assessed all five dimensions; most surveys covered aspects of one specific dimension for example, financial toxicity as part of the Affordability component, patient–provider communication as part of the Acceptability component, but these were not comprehensive even within each dimension; (2) none of the existing instruments appeared to have considered the varying experiences of different racial groups in developing the items, making it unclear if the questions would capture the experiences of black and Hispanic patients, the major focus of ORCHiD and (3) combining questions or subscales from various surveys into one HCA instrument would likely result in limited validity with poor psychometric properties. Therefore, we addressed these critical limitations by conducting an initial qualitative study that included focus group discussions and cognitive interviews with cancer survivors to provide content validation data for the HCA survey to be used in our prospective cohort. Additionally, the original framework was developed in the context of primary care. Data gained from focus groups identified how each HCA dimension applied to a socioeconomic and racially diverse group of patients with cancer, allowing us to obtain measures of access specific to cancer care in each of the dimensions. The process of concept elicitation with focus groups, cognitive interviews and pilot testing is described below and summarised in figure 3.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Conceptual framework for qualitative study and survey development.

1. Draft survey: We began drafting a questionnaire by pulling survey items from existing, validated surveys that addressed the topics of interest. Validated questionnaires are instruments that have undergone evaluation to ensure the instrument measures what it is intended to measure.

2. Concept elicitation: To guide the development of survey questions that were not captured by existing instruments, we conducted focus group discussions with 32 female cancer survivors (63% black and 18% Hispanic) using a semistructured interview guide. The focus groups solicited input regarding their perceptions of each dimension of HCA, experiences accessing care after their cancer diagnosis, and facilitators and barriers to receiving quality treatment. Information from the focus groups were synthesised and analysed and used to guide survey development. Focus group participants were provided a US$25 gift card for participation.

3. Cognitive testing: We conducted cognitive testing of the HCA survey items with in-depth interviews among seven patients with OC diagnosed at Duke University (three black, one Hispanic and three white patients). Interview participants were asked to answer the questionnaire, think about their responses (i.e., retrospective probing) and describe their thought processes after responding. The interviewer probed on (1) comprehension (level of understanding), (2) clarity (level of straightforward meaning), (3) knowledge and memory (ease of recall of information needed to respond) and (4) judgement (ease of fitting personal experiences to the measure response options). Debriefing forms were completed for each interview and used by the study team to review findings and modify questions as needed. Participants were provided a US$25 gift card for participation.

4. Pilot testing: With the ORCHiD survey at near final version, having been edited for length, clarity and content, we conducted a pilot study of 50 patients with OC recruited from the Duke Cancer Center and the University of Alabama at Birmingham Comprehensive Cancer Center. The full-length survey was administered by phone (preferred option) or self-administered electronically. A small proportion of patients preferred to complete the paper version and were mailed a shortened version of the full survey to enhance completion. The biospecimen substudy component was also tested in this phase. Feedback from pilot study participants regarding the survey was incorporated into the draft version before being finalised for the main study. Participants were provided a US$25 gift card incentive for completing the survey and an additional US$20 gift card for providing biospecimen.

Measuring guideline-recommended treatment

In both the retrospective and prospective cohorts, we will use previously well-validated treatment algorithms combining registry and claims data to assess receipt of guideline-adherent OC treatments.17–19 We will compare treatments received to the treatments recommended in professional guidelines published by the National Comprehensive Cancer Network.20 We will also select well-defined supportive care quality measures from the CMS Quality Payment Programme, including medications for treatment of pain, depression or anxiety. We will evaluate care quality for comorbid conditions, measured by hospitalisations and emergency room visits for potentially avoidable admissions related to suboptimal management of multimorbidity. Concordance between recommended treatment and treatment outcomes in this cohort will be assessed from multiple overlapping data sources: (1) patient-reported treatment information from the survey (in prospective cohort); there is evidence that patients can reliably report receipt of recent radiation and chemotherapy, although not the specific agents used,21–24 (2) cancer registry data on surgery and chemotherapy for all participants; these data have been shown to have a level of reliability similar to claims data,25 (3) claims data on treatment through linkages with Medicare or Medicaid claims data and (4) a manual review of medical records for patients in the Kentucky registry as further validation. These multiple layers of treatment information in the primary survey cohort will minimise the likelihood of misclassification and also enable us to assess the validity of treatment information obtained from survey data.

Measuring PROs and other patient variables

We will measure patient-reported symptoms using the Patient-Reported Outcomes Measurement Information System (PROMIS) four-item short forms in the prospective cohort study. PROMIS item banks have been extensively tested,26 27 and short forms are reliable and valid for use in the general US population, in cancer-specific populations28 and in non-white and low-income subgroups.29–32 We will assess four symptom domains widely recognised as prevalent and manageable in cancer survivors: fatigue, depression, anxiety and sleep disturbance. Self-efficacy for healthcare will also be assessed with a PROMIS measure.33 34 We will use two out of five possible subdomains from this item bank: confidence in managing symptoms and confidence in managing medications and also assess functional status. Registry-derived data including sociodemographics, month and year of diagnosis, cancer site, histologic type, stage at diagnosis, survival time and vital status, and first course of therapy will be obtained for all patients in both cohorts. Since comorbidity status is a significant predictor of poor survival with marked racial differences, we will assess comorbidities in both cohorts. In the retrospective SEER-Medicare cohort, we use claims data to create a comorbidity index,35 and in the prospective cohort, we ask respondents about common and serious comorbid conditions adapted from the Charlson index. We also include measures of select risk factors (e.g., smoking, body mass index) that are both associated with poor survival and also potentially modifiable targets for improving outcomes.36 37

Biospecimen collection

Following completion of the survey in the prospective cohort, participants will be asked if they are interested in participating in future research. If they decline, no further studies will be introduced, however, if they agree, the participants will be offered the opportunity to participate in the biospecimen substudy. Interested participants will be asked to provide self-collected cervicovaginal microbiome specimen using the DNA Genotek OMR-130 kit (or similar) and saliva sample using DNA Genotek OGR-500 kit (or similar). Samples will be shipped back to the study team at Duke for processing and analysis, and molecular data will be linked to clinical data obtained in the ORCHiD study for future analysis.

Patient and public involvement

Essentially, the goal of ORCHiD is to address these knowledge gaps by (1) fully characterising HCA dimensions among black, Hispanic, and white patients with OC, (2) evaluating the association of HCA dimensions with racial disparities in receipt of quality, guideline-adherent initial treatment, supportive care and survival and (3) and characterising the biological mechanisms underlying OC disparities. Patients, cancer survivors and community members were involved in the study design, and contributed important insights into the recruitment and conduct of this research study. The study protocol was presented to the Duke Cancer Institute Community Advisory Council for feedback, and information obtained during the focus group with cancer survivors and the pilot study of ovarian cancer survivors were incorporated into the study design. Feedback on each step of the study design, pilot and ongoing recruitment activities are disseminated back to patients and survivors, with a plan to provide ongoing updates.

Analytical plan

Descriptive statistics will be conducted for all variables in each study cohort overall and by race, region and age group. We will compare categorical variables by group using χ² tests, and continuous variables using t-tests or non-parametric tests as appropriate. Hierarchical multiple regression analysis will be used to evaluate the association between each main exposures of interest (HCA dimensions) and study outcomes (guideline-adherent treatment, PROs, survival), accounting for the clustering of patients within neighbourhoods (e.g., county), physicians and hospitals, and adjusting for a wide range of relevant study covariates. Variable collinearity will be assessed using the variance inflation factor (VIF) method. Covariates with VIF values of 10 or higher will be dropped from the multivariable models. Additionally, the planned factor analysis will be used to create composite scores for the HCA dimensions; these scores will be used in lieu of adjustment for each individual variable, which should help to reduce issues of collinearity between variables of interest. Non-response from the survey and its characteristics will be recorded and evaluated to identify potential biases. We will conduct multiple imputation, for instance, full conditional specification approach, to handle missing data. If missing data are deemed to be not missing at random, non-standard conditionally specified models will be used.

Sample size and power analysis

To estimate the sample size for the primary survey cohort, we focused our power calculation on the outcome for Aim 2: receipt of guideline-adherent treatment. A 10%–14% difference in the receipt of guideline-adherent treatment was observed between black and white patients ages 65+ in a population-based cohort,2 38 with larger differences observed in those less than age 65.4 39 We will recruit all potential black patients with OC from participating registries, age group-matched white patients and all potential Hispanic patients with OC from Texas and California registries. We will, therefore, have a total sample size of 9744 patients in the SEER-Medicare cohort, and 1641 patients (413 blacks, 299 Hispanics and 929 whites) in the primary survey cohort. Accounting for intraclass correlation coefficient (conservatively assumed to not exceed 0.1) within the multilevel statistical model, and assuming 60% of study subjects reside in neighbourhoods with high HCA availability, we will have 80% power to detect racial differences in treatment as small as 6% using SEER-Medicare data and as small as 14% using the primary survey cohort data. Based on similar assumptions, we will be well powered to detect hazard ratios as small as 1.09 in SEER-Medicare data, and as small as 1.21 in the primary survey cohort. We will also have 80% power to detect at least a 1.5-point difference (or 0.15*SD) in PROMIS scores.