We used two data sources. To identify patients with terminal illnesses and the physicians who provided outpatient care, we used the Colorado All-Payer Claims Database (APCD). The APCD is a statewide, comprehensive administrative record that includes inpatient, outpatient, physician, and facility entitlements for nearly all patients receiving care in Colorado. Reporting is required for all insurance companies and plans except for federal health care facilities (e.g., Veterans Health Administration hospitals) and self-insured group health plans. So it’s not a voluntary effort that could lead to bias in reporting. The APCD also includes beneficiary demographics, including age and gender, insurance carrier, and hospital identifiers, but does not contain reliable data on race/ethnicity. This data can be purchased by researchers. To obtain additional information on physician characteristics, we used commercially available data from IQVIA. This data links National Provider Identifiers (NPIs, available from the APCD) to information such as specialty physician, practice location, and mailing address.
Creation of a patient cohort
We attempted to identify a patient cohort resembling those patients known to have filled MAID prescriptions in Colorado in 2017-2018. Publicly available reports from the Colorado Department of Public Health and Environment (CDPHE) show that of the 193 patients who filled a MAID prescription in 2017-2018, the most common conditions were malignant neoplasms, progressive neurodegenerative diseases (such as amyotrophic lateral sclerosis, and progressive supranuclear palsy), chronic lower respiratory disease (eg, chronic obstructive pulmonary disease), and cardiac disease (eg, congestive heart failure)11. The CDPHE also reports that more than 75% of patients receiving a MAID prescription were enrolled in a hospice. Therefore, the inclusion criteria of our patient cohort included (1) a diagnosis listed above and (2) receipt of hospice services.
We used International Classification of Diseases (tenth revision) (ICD-10) codes to identify patients with these diagnoses, and Current Procedural Technology (CPT) and Healthcare Common Procedure Coding System (HCPCS) codes to identify patients who have used hospice services (supplementary table). . Among all patients who sought health care in Colorado from 2017-2018, we identified those with ICD-10 primary diagnosis codes for inpatient or two outpatient claims in 2017-2018 for the conditions listed in the CDPHE report. Among these patients, we then identified those receiving hospice services using CPT and HCPCS codes for hospice care planning or modifiers for services provided by hospice care.
Formation of a doctor cohort
We then identified a cohort of physicians who routinely provided outpatient care to the patient cohort described above. We hypothesized that opportunities for patients to ask about MAID would occur most frequently during outpatient clinic visits. Therefore, in the cohort of patients described above, we identified outpatient clinic visits using CPT codes 99201–99205 (new patient visits) and codes 99211–99215 (repeat patient visits). We used National Provider Identifier (NPI) numbers to identify the physicians and providers for each of these visits. We included both individual physician NPIs and organizational NPIs (ie group practices). We used IQVIA data to identify individual physicians practicing within each organizational NPI. We excluded individual NPIs from advanced practice providers (e.g., nurses and physician assistants) because Colorado law does not allow these physicians to prescribe MAID.
We sent out three survey waves (200 surveys in the first two waves, 183 surveys in the third wave; 583 surveys in total). For each survey wave, we ranked physicians in order of their estimation based on their likelihood of being asked about MAID (stratified probability sampling), based on the research team’s experience and responses from the first and second survey waves. A doctor could only be questioned once; If they were surveyed in a previous wave, they were excluded from the ranking process in subsequent waves.
For the first wave, the classification criteria consisted of two elements: the patient-provider ratio and the specialty of the practice. For patient-to-provider ratio, we assigned physicians and organizations points based on the number of patients in our cohort they saw (e.g., a single physician treating 2 patients in the cohort had a 2 and an organization with 5 physicians and saw 10 patients also had a ratio of 2). Physicians or organizations with a patient-to-provider ratio >3 received 3 points; 2–3 received 2 points; 1 received 1 point; and < 1 received 0 points. Physicians practicing within the same organization received the same score for this criterion because organizational NPI could not be attributed to individual physicians practicing under the organizational NPI. 3 points were awarded for the practice focus areas of medical haematology/oncology, hospice and palliative medicine and geriatrics; Neurology, family medicine and internal medicine received 2 points; Pulmonary and cardiovascular medicine received 1 point; and all others received 0 points. This was based on CDPHE descriptive reports of the diagnoses of patients prescribed MAID. We then surveyed the top 100 physicians in each of the individual and organizational NPI groups (total n=200). The surveys were sent in two color codes, one for physicians with individual NPIs and one for physicians with organizational NPIs.
For the second wave, we varied the placement criteria for the specialty practice and added a placement criteria based on the placement of the practice. This was based on results from the first wave, which indicated that the vast majority of physicians participating in MAID activities provided both inpatient and outpatient care rather than just one facility. Specialists in medical haematology/oncology, hospice and palliative medicine and geriatrics received 2 points; Neurology, family medicine and internal medicine received 1 point; and all others received 0 points. Physicians who bill both outpatients and inpatients receive 1 point; those who scored in only one setting received 0 points. The ranking criteria for the patient-to-provider ratio were not changed. We then surveyed the top 100 physicians in each of the individual and organizational NPI groups (total n=200). As in wave 1, surveys were sent in two color codes based on the use of individual or organizational NPI. For the third wave, we only surveyed the physicians who billed according to their own individual NPI and did not change any other ranking criteria from wave 2. This was based on results from the first and second waves, which indicated that physicians who billed according to their own individual NPI were more likely to have discussed MAID with their patients. In the third wave, all remaining physicians in the physician cohort were interviewed with individual NPIs (n = 183).
It is important to highlight how survey administration differed from standard practice. Firstly, given the sensitivity of the topic, we conducted a completely anonymous survey, meaning there was no way of attributing responses to individual doctors. Second, we asked a very limited number of demographic questions with exceptionally broad response categories to provide additional reassurance to respondents that they could not be identified from their survey responses. Third, we used a very short survey (4 pages) that could be completed in less than 15 minutes. Fourth, instead of providing a check or prepaid gift card, we provided a $50 cash incentive upfront, as using these mechanisms or incentives would require identifying respondents to the research team after the survey was completed. Finally, we did not undertake follow-up activities such as phone calls and additional mailings as we were not able to identify those who had responded to the survey and those who had not.
In the survey, we measured the extent to which physicians were willing, willing, or actually speaking to patients about MAID, referring patients for MAID, serving as MAID counselors, and as MAID caregivers. Due to the extreme sensitivity surrounding MAID, and to ensure physicians’ complete anonymity, we have a very limited number of demographic questions with intentionally broad response categories related to gender, specialty, length of medical practice, race/ethnicity, and personality characteristics posed exercise setting. Respondents were also instructed to freely skip questions they did not wish to answer.
We used standard descriptive statistics (t-tests or chi-square tests) to assess the differences in the characteristics of the physicians surveyed and the self-reported characteristics of the survey participants. We used chi-square tests to identify differences in the proportions of survey respondents participating in MAID activities. Response rates were calculated according to the American Association for Public Opinion Research (AAPOR) Standard Definition Version 4.112. The study was approved by the Colorado Multiple Institutional Review Board. All investigations were carried out in accordance with the relevant guidelines/regulations. Because the survey was completely anonymous, the Colorado Multiple Institutional Review Board considered participation in the survey to be consent. The survey was conducted by traditional mail. The design and objectives of the study were explained to the participants. Respondents were also informed that the poll results would be made public.
Ethics Approval and Consent
The study was approved by the Colorado Multiple Institutional Review Board.