Association between physicians' geriatric training and patterns of end‐of‐life care delivered to persons with dementia

Geriatric training is designed to prepare physicians to meet the complex needs of older adults, including persons with dementia at the end‐of‐life (EOL) stage. We sought to compare patterns of EOL care delivered to persons with dementia between physicians with versus without geriatric training.


INTRODUCTION
Dementia is a progressive, life-limiting illness with significant morbidity and mortality.The expected number of persons with Alzheimer's disease is projected to more than double from 6.5 million to 12.7 million by 2050, 1 and is currently the sixth-leading cause of death among adults. 2][9][10][11] There are currently 24 subspecialty (fellowship) training programs in Internal and Family Medicine. 12,13Among them, subspecialty training in geriatric medicine is designed to prepare physicians to meet the complex care needs of older adults, including persons with dementia at the last stage of their lives. 146][17] However, little is known as to the association between physicians' geriatric training and patterns of EOL care delivered to persons with dementia.Given the projected increase in persons with dementia and concerns about the quality of EOL care received by persons with dementia, it may be useful to understand whether patterns of EOL care differ between physicians with versus without formal geriatric training.
To address this knowledge gap, we analyzed the nationally representative data of Medicare beneficiaries and examined the association between geriatric training and patterns of EOL care among persons with dementia using a broad range of measures.We defined geriatric training as a self-reported specialty of geriatric medicine and/or board certification in geriatric medicine.We also investigated whether physicians with more clinical experience with dementia patients provide different patterns of EOL care in dementia to understand the mechanism of the association between geriatric training and patterns of EOL care.

Data source and study participants
We linked three databases: (1) 20% of Medicare claims data 2016-2018, (2) long-term care Minimum Data Set (MDS) 3.0 2016-2018, (3) Medicare Data on Provider Practice and Specialty (MD-PPAS) data, and (4) Doximity data.The Medicare claims data provide beneficiary characteristics including age, verified death dates (available for more than 99% of decedent beneficiaries), monthly fee-for-service coverage status, and indicators for chronic conditions based on definitions by the Chronic Condition Data Warehouse, including dementia (i.e., Alzheimer's Disease and Related Disorders or Senile Dementia). 18The MDS is the dataset of a federally mandated clinical assessment of all residents in Medicare-or Medicaid-certified nursing homes, and we used MDS to identify long-term nursing home residents.The MD-PPAS data includes physician characteristics, such as gender, specialty, and location.Specialty in the MD-PASS data is extracted from the Key points

Why does this paper matter?
Provision of geriatric training for physicians may have the potential to improve the quality of endof-life care delivered to persons with dementia.
self-reported specialty information in the Provider Enrollment and Chain/Ownership System (PECOS), the national enrollment system for Medicare clinicians and suppliers. 19Doximity is a company providing online professional networking services for US physicians, and it provides data on physician characteristics for both registered members of the service and nonregistered physicians through multiple sources and data partnerships including the American Board of Medical Specialties. 20,21e first identified beneficiaries 66 years and older with dementia who died in 2016-2018 with fee-for-service coverage during the last 6 months of life.We then attributed each beneficiary's outcome to a physician who provided more than 50% of primary care service evaluationand-management (E&M) claims in the last 6 months of life (hospice billing codes are not included; see Supplementary Table S1 for included billing codes).3][24] We restricted physicians to those with primary care specialties based on the practitioner taxonomy codes in the MD-PPAS (family practice, general practice, internal medicine, or geriatric medicine). 25We excluded beneficiaries (1) who died before July 1, 2016, because we used the information during the last 6 months of life to attribute beneficiaries to physicians; (2) who were long-term nursing home residents (defined as those who had a comprehensive or quarterly MDS assessment within 90 days before death 26 ) because endof-life care significantly differs by nursing home status 27 ; (3) whose outcomes were unable to be attributed to a physician (e.g., a beneficiary did not receive primary care services during the last 6 months of life); or (4) who have missing data on important variables.See Figure 1 for the study participant flow chart.

Geriatric training and experience in dementia care
Our primary exposure variable was a binary indicator of whether a physician to whom a beneficiary's outcomes were attributed had geriatric medicine training.We defined geriatrics-trained physicians as (1) those with a self-reported specialty of geriatric medicine in the MD-PASS data and/or (2) those with board certification in geriatric medicine (both active and inactive) through the American Board of Internal Medicine (ABIM) based on the Doximity data.Some physicians were certified in geriatrics without undergoing geriatrics fellowship training during the so-called "grandfather period" (1988-1994) 28 ; therefore, we conducted a sensitivity analysis by restricting our sample to those who completed residency after this period (see Statistical analysis section for more details).Our secondary exposure variable was physicians' experience in dementia care.We categorized physicians into quartiles based on the proportion of beneficiaries with dementia under care and included it as an ordinal variable for quartiles.

Patterns of end-of-life care
We examined patterns of EOL care among persons with dementia in two areas: (1) advance care planning (ACP) and palliative care, and (2) high-intensity EOL care, using Medicare fee-for-service claims data.We measured advance care planning and palliative care using three measures: (i) receipt of advance care planning, (ii) receipt of palliative care counseling, and (iii) hospice enrollment in the last 90 days of life (see Supplementary Table S2 for specific Current Procedural Terminology [CPT] codes and International Classification of Disease [ICD]-10 codes used to define these study outcomes).We examined highintensity EOL care by whether a beneficiary received any of the following six healthcare services in the last 30 days of life: (i) emergency department visits, (ii) hospital admissions, (iii) intensive care unit (ICU) admissions, (iv) mechanical ventilation, (v) cardiopulmonary resuscitation (and/or defibrillation), or (vi) placement of feeding tubes (see Supplementary Table S2 for included CPT codes and ICD-10 codes).

Adjustment variables
We adjusted for multiple beneficiary characteristics: age at the time of death (continuous), sex, race and ethnicity (non-Hispanic White, non-Hispanic Black, Hispanic, or Other), coexisting conditions (dummy variables for 25 Chronic Condition Data Warehouse chronic conditions-excluding two dementia-related conditions), median annual household income estimated from residential zip codes (categorized into quintiles), and Medicaid coverage.We also adjusted for fixed effects for year of death and Hospital Service Areas (HSAs) to account for secular trends and geographic variation (effectively comparing physicians with and without geriatric training practicing in the same year and HSA).We also included physician sex and years of clinical experience (1-10, 11-20, 21-30, or ≥31 years) based on MD-PASS data as adjustment variables.We estimated years of clinical experience by subtracting 3 years (the duration of Internal and Family Medicine Residency programs) from years since graduation from medical school.

Statistical analysis
First, we examined the association between physicians' geriatric training and each measure for patterns of EOL care in dementia by fitting multivariable linear regression models adjusting for beneficiary and physician characteristics as well as year of death and HSA fixed effects.The unit of analysis was beneficiary, and we clustered standard errors at the physician level to account for potential correlation among beneficiaries cared for by the same physician.We also used the Benjamini-Hochberg method to account for the multiple comparisons across outcomes, and report both unadjusted and adjusted p-values (an adjusted p-value of less than 0.05 was considered statistically significant). 29,30We used linear regression models for the interpretability of the regression coefficients (i.e., linear probability models). 31econd, to examine whether physicians' experience in dementia care is associated with patterns of EOL care in dementia, we fit multivariable linear regression models similar to the main analysis but including an ordinal variable for quartiles for the proportion of beneficiaries with dementia under care, instead of the indicator for physicians' geriatric training status, to evaluate the trend across quartiles.
Statistical analyses were conducted using SAS version 9.4 and Stata/MP 16.1.The institutional review boards at the University of California, Los Angeles reviewed the study and waived informed consent.

Sensitivity analyses
First, we conducted a sensitivity analysis by restricting our sample to those who were not enrolled in hospice 30 days before their deaths, because hospice enrollees might have been cared for by a hospice medical director with a geriatrics background in addition to a physician whom a beneficiary is attributed to.Second, we excluded physicians who graduated from medical schools before 1992 (i.e., completed Internal or Family Medicine residency before 1995) from the analysis, because this group might have been certified in geriatrics without undergoing geriatrics fellowship training during the grandfather period (1988-1994) and different from those who underwent geriatrics fellowship training. 28astly, we included time from dementia diagnosis to death (categories of <4 years, 4-6 years, or ≥6 years 32 ) as an additional adjustment variable in the models as a proxy for the severity of dementia, because the severity of dementia can affect the association between physicians' geriatric training and patterns of EOL care.

Physician and beneficiary characteristics
We included 99,631 beneficiaries with dementia who died in 2016-2018 (see Figure 1 for the study participant flow chart).Among them, 8340 (8.4%) beneficiaries were attributed to physicians with geriatric training (n = 1546), and 91,291 (96.2%) were attributed to physicians without geriatric training (n = 24,451).Physicians with geriatric training were younger, more likely to be female, and had less clinical experience (Table 1).Beneficiaries under the care of physicians with geriatric training were more likely to be old, female, non-Hispanic White, living in zip codes with higher median household incomes, and have fewer coexisting conditions, compared to those under the care of physicians without geriatric training (Table 1).

Physicians' geriatric training and patterns of EOL care
We found that beneficiaries with dementia under the care of physicians with geriatric training had a higher proportion of advance care planning (adjusted proportion, 15.8% vs. 13.0%,adjusted difference, 2.8 percentage points [pp, 95% CI, 1.7-3.9]),palliative care counseling (22.4% vs. 20.9%,1.4 pp [0.3-2.5]), and hospice enrollment in the last 90 days of life (63.7% vs. 60.6%,3.2 pp [2.0-4.3])compared to those under the care of physicians without geriatric training (Figure 2 and Supplementary Table S3).We also found that physicians '

Clinical experience in dementia care and patterns of EOL care
We found that beneficiaries under the care of physicians with more experience in dementia care (i.e., a higher proportion of beneficiaries with dementia under care) had a higher proportion of advance care planning and hospice enrollment in the last 90 days of life (both p-fortrend <0.001) (Table 2).We also found that beneficiaries under the care of physicians with more experience in dementia care had a lower proportion of emergency department visits, hospital admissions, ICU admissions, use of mechanical ventilation ( p-for-trend <0.02), and use of cardiopulmonary resuscitation (all p-for-trend <0.001 unless otherwise noted) in the last 30 days of life.

Sensitivity analyses
Our sensitivity analyses by restricting our sample to non-hospice enrollees (Table 3), by excluding physicians who might have been certified in geriatrics without having geriatrics fellowship training (Supplementary Table S4), and by including time from dementia diagnosis to death as an additional adjustment variable (Supplementary Table S5) yielded similar results to the main analysis.

DISCUSSION
Using nationally representative data on Medicare feefor-service beneficiaries with dementia, we found that persons with dementia who were cared for by physicians with geriatric training received less intensive EOL care, including ED visits, hospital admissions, ICU admissions, and mechanical ventilation and more hospice use compared to those without geriatric training.4][35] Our data support this in that patients cared for by a physician with geriatric training were more likely to have opportunities to express their preferences with their team as evidenced by more frequent use of advance care planning and palliative care counseling.Our findings should be informative for policymakers, educators, and patients because they suggest the potential positive effects of physicians' geriatric training to change the patterns of EOL care delivered to persons with dementia.While the observed differences between the two groups may seem modest (2-4 percentage points), these differences are arguably meaningful as our data suggest that more than 800,000 persons die with dementia every year in the United States.
Geriatric training programs grew considerably after 1978 when the Institute of Medicine (IOM) called for  42,43 Given our findings, providing more geriatric training during or after residency (e.g., through a creative curriculum, 44 and a Continuing Medical Education requirement 45 ) or increasing (both financial and nonfinancial) incentives for physicians to undergo a formal geriatric fellowship program has the potential to improve the quality of EOL care delivered to persons with dementia.Our finding of the dose-response association between the proportion of patients with dementia under care and patterns of EOL care additionally supports the potential benefits of training and exposure.We observed no evidence that the placement of feeding tubes in the last 30 days of life differs between beneficiaries with dementia under the care of physicians with versus without geriatric training.This may be because the current guidelines provide sufficient guidance and have been successful in influencing physicians' practice patterns for these healthcare services regarding when these invasive measures align with patients' care preferences (e.g., Choosing Wisely 46 ).
Our study builds upon previous work that examined the association between physicians' geriatric training and the quality of care.One study in a tertiary care hospital found that geriatrician-hospitalists provided earlier physical therapy, shorter duration of indwelling bladder catheters, and less frequent use of benzodiazepines or anticholinergics for hospitalized older adults, compared to hospitalists without geriatric training. 16Another study conducted at three primary care clinics of an academic medical center also showed that geriatricians were more likely to avoid potentially inappropriate medications and screen for geriatric syndromes compared to generalists. 15imilarly, a case-control study at a single medical center in Taiwan found that patients who received care at a geriatric outpatient clinic had fewer emergency department visits and hospitalizations compared to those at a nongeriatric outpatient clinic. 17However, these studies did not examine patterns of EOL care for persons living with dementia.We provide new evidence on the association between geriatric training and a wide range of EOL care measures for persons with dementia using a US nationally representative sample.
Our study has limitations.8][49] Similarly, because our outcomes relied on claims data, they can be influenced by the coding practice.For example, the low prevalence of ACP we observed is likely due to the low uptake of ACP billing codes that were newly implemented in 2016.Second, we were unable to determine the severity of dementia, although our sensitivity analysis by including time from dementia diagnosis to death as an additional adjustment variable as a proxy for the severity of dementia yielded similar findings.Third, our physician attribution algorithm may lead to incorrect attributions for some beneficiaries.Similarly, because our Doximity data did not contain information from the American Board of Family Medicine (ABFM), which provides board certification in geriatric medicine along with the ABIM, we may have misclassification of geriatric training.However, we additionally used the MD-PPAS data to supplement the Doximity data, and this potential misclassification should lead to conservative estimates.Fourth, it is possible that unobserved factors are confounding the association between physicians' geriatric training and the patterns of EOL care in dementia.For example, patients and their caregivers who prefer less intensive EOL care might seek care from a geriatrician, rather than a general practitioner.Fifth, our study was not able to determine whether highintensity EOL care was or was not concordant with patients' goals.While previous studies have suggested that persons with dementia and their caregivers may desire less aggressive care as their disease progresses, 6,[33][34][35] high-intensity EOL care might have been appropriate for some patients.Lastly, our findings may not be generalizable to other populations, such as younger populations and Medicare Advantage enrollees.
In summary, using a national dataset of decedents with dementia, we found that those cared for by physicians with geriatric training received more advance care planning and palliative care and less intensive EOL care.Our findings suggest that interventions for physicians without formal geriatric training who care for persons with dementia might need to be considered, if evidence emerges that short-duration training is effective in improving the quality of EOL care.

AUTHOR CONTRIBUTIONS
Study concept and design, interpretation of data, and preparation of the manuscript: Hiroshi Gotanda and Yusuke Tsugawa.Interpretation of data and preparation of the manuscript: Jessica J. Zhang.Interpretation of data and revision of the manuscript critically for important intellectual content: David B Reuben and Anne M Walling.Acquisition of data, analysis of data, and revision of the manuscript critically for important intellectual content: Haiyong Xu.Acquisition of data and revision of the manuscript critically for important intellectual content: Anupam B. Jena and Nate Gross.

F I G U R E 2
Association between physicians' geriatric training status and patterns of end-of-life care.Bars indicate the adjusted percentages of decedents 66 years and older with dementia who experienced each outcome by physicians' geriatric training status based on 20% of Medicare claims data 2016-2018.We fit linear regression models adjusted for the characteristics of beneficiaries (age, sex, race and ethnicity, 25 coexisting conditions, zip-code level median annual household income, and Medicaid coverage) and physicians (sex and years of clinical experience).We also include fixed effects for year of death and Hospital Service Areas in the models.See the main text for the algorithm to attribute each beneficiary's outcomes to physicians and the definition of physicians' geriatric training.p-Values are adjusted with the Benjamini-Hochberg method to account for the multiple comparisons across outcomes (an adjusted p-value of less than 0.05 is considered statistically significant).ED, emergency department; ICU, intensive care unit.
Physician and beneficiary characteristics by physicians' geriatric training status.
T A B L E 1Note: Data are for decedents 66 years and older with dementia from 20% of Medicare claims data 2016-2018.
geriatric training was associated with fewer beneficiaries with dementia having emergency department visits (55.1% vs. 59.1%,À4.0 pp [À5.1 to À2.8]), hospital admissions (48.8% vs. 52.3%,À3.5 pp [À4.6 to À2.4]), ICU admissions (24.9% vs. 27.4%,À2.5 pp T A B L E 2 Association between the proportion of beneficiaries with dementia under care (quartiles) and patterns of end-of-life care.We categorized physicians into quartiles based on the proportion of beneficiaries with dementia under care and included it as an ordinal variable for quartile, instead of the indicator for physicians' geriatric training status, in the multivariable linear regression models similar to the main analysis.See the main text for more details.
T A B L E 3 Association between physicians' geriatric training status and high-intensity end-of-life care among non-hospice enrollees.Using decedents 66 years and older with dementia who were not enrolled in hospice 30 days before their deaths from 20% of Medicare claims data 2016-2018, we fit linear regression models adjusted for the characteristics of beneficiaries (age, sex, race and ethnicity, 25 coexisting conditions, zip-code level median annual household income, and Medicaid coverage) and physicians (sex and years of clinical experience).We also include fixed effects for year of death and Hospital Service Areas.See the main text for more details. Note: