Association between physician age and patterns of end-of-life care among older Americans.

BACKGROUND: End-of-life (EOL) care patterns may differ by physician age given differences in how physicians are trained or changes associated with aging. We sought to compare patterns of EOL care delivered to older Americans according to physician age. METHODS: We conducted a cross-sectional study of a 20% sample of Medicare fee-for-service beneficiaries aged ≥66 years who died in 2016-2019 (n = 487,293). We attributed beneficiaries to the physician who had >50% of primary care visits during the last 6 months of life. We compared beneficiary-level outcomes by physician age (<40, 40-49, 50-59, or ≥60) in two areas: (1) advance care planning (ACP) and palliative care; and (2) high-intensity care at the EOL. RESULTS: Beneficiaries attributed to younger physicians had slightly higher proportions of billed ACP (adjusted proportions, 17.1%, 16.1%, 15.5%, and 14.0% for physicians aged <40, 40-49, 50-59, and ≥60, respectively; p-for-trend adjusted for multiple comparisons <0.001) and palliative care counseling or hospice use in the last 180 days of life (64.5%, 63.6%, 61.9%, and 60.8%; p-for-trend <0.001). Similarly, physicians' younger age was associated with slightly lower proportions of emergency department visits (57.4%, 57.0%, 57.4%, and 58.1%; p-for-trend <0.001), hospital admissions (51.2%, 51.1%, 51.4%, and 52.1%; p-for-trend <0.001), intensive care unit admissions (27.8%, 27.9%, 28.2%, and 28.3%; p-for-trend = 0.03), or mechanical ventilation or cardiopulmonary resuscitation (14.2, 14.9%, 15.2%, and 15.3%; p-for-trend <0.001) in the last 30 days of life, and in-hospital death (20.2%, 20.6%, 21.3%, and 21.5%; p-for-trend <0.001). CONCLUSIONS: We found that differences in patterns of EOL care between beneficiaries cared for by younger and older physicians were small, and thus, not clinically meaningful. Future research is warranted to understand the factors that can influence patterns of EOL care provided by physicians, including initial and continuing medical education.

Association between patient-surgeon gender concordance and mortality after surgery in the United States: retrospective observational study.

OBJECTIVE: To determine whether patient-surgeon gender concordance is associated with mortality of patients after surgery in the United States. DESIGN: Retrospective observational study. SETTING: Acute care hospitals in the US. PARTICIPANTS: 100% of Medicare fee-for-service beneficiaries aged 65-99 years who had one of 14 major elective or non-elective (emergent or urgent) surgeries in 2016-19. MAIN OUTCOME MEASURES: Mortality after surgery, defined as death within 30 days of the operation. Adjustments were made for patient and surgeon characteristics and hospital fixed effects (effectively comparing patients within the same hospital). RESULTS: Among 2 902 756 patients who had surgery, 1 287 845 (44.4%) had operations done by surgeons of the same gender (1 201 712 (41.4%) male patient and male surgeon, 86 133 (3.0%) female patient and female surgeon) and 1 614 911 (55.6%) were by surgeons of different gender (52 944 (1.8%) male patient and female surgeon, 1 561 967 (53.8%) female patient and male surgeon). Adjusted 30 day mortality after surgery was 2.0% for male patient-male surgeon dyads, 1.7% for male patient-female surgeon dyads, 1.5% for female patient-male surgeon dyads, and 1.3% for female patient-female surgeon dyads. Patient-surgeon gender concordance was associated with a slightly lower mortality for female patients (adjusted risk difference -0.2 percentage point (95% confidence interval -0.3 to -0.1); P<0.001), but a higher mortality for male patients (0.3 (0.2 to 0.5); P<0.001) for elective procedures, although the difference was small and not clinically meaningful. No evidence suggests that operative mortality differed by patient-surgeon gender concordance for non-elective procedures. CONCLUSIONS: Post-operative mortality rates were similar (ie, the difference was small and not clinically meaningful) among the four types of patient-surgeon gender dyads.

Effect of age-based left-digit bias on stroke diagnosis: Regression discontinuity design.

BACKGROUND: Little is known about how left-digit bias- where humans tend to discretely categorize their decisions based on the left-most digit of the continuous variables-based on patients' age affects the initial diagnosis of stroke patients. The aim of this study is to examine whether there is a discontinuous change in the ordering of imaging tests for stroke at the age threshold of 40 years old, which is indicative of left-digit bias, and whether the effect varies by patient sex. METHODS: We conducted a cohort study by regression discontinuity design (RDD). We analyzed the claims database from a nationwide working-age health insurance plan in Japan. Patients who had after-hours hospital visits from January 2014 through December 2019 were included in our analysis. The exposure is patients' age, and the primary outcome was physicians' ordering imaging tests (CT or MRI) to diagnose stroke during the index visit. RESULTS: Of 293,390 total visits, 48,598 visits within data-driven optimal bandwidths of 6.0 years from the cut-off of 40 years were included for the RDD analysis (mean age 40.8 years [standard deviation 3.4], female 50.5%). The baseline probability of receiving imaging tests for stroke diagnosis was 0.9%. Physicians had a higher likelihood of ordering imaging tests when patients' age was above 40 years compared with when patients' age was just below 40 years (adjusted difference, +0.51 percentage points [pp], 95%CI, +0.13 to +1.07 pp; P = 0.01). We found a significant discontinuous change in the ordering of imaging tests for stroke at 40 years for male patients (+0.84 pp, 95%CI, +0.24 to +1.69 pp; P = 0.009) but not for female patients. CONCLUSIONS: Physicians have a cognitive bias in estimating stroke risk and, consistent with a left-digit bias, treat male patients aged 40 and just below differently. This pattern was observed only among male patients.

Association of Education With Dementia Incidence Stratified by Ethnicity and Nativity in a Cohort of Older Asian American Individuals.

Importance: High education protects against dementia, but returns on educational attainment may be different across sociodemographic groups owing to various social factors. Asian American individuals are a growing and diverse group, but little research has assessed dementia determinants in this population. Objective: To examine the association of education with dementia in a large cohort of Asian American individuals, stratifying by ethnicity and nativity. Design, Setting, and Participants: This cohort study used electronic health record (EHR) and survey data from the Research Program on Genes, Environment, and Health and the California Men's Health Study surveys (2002-2020). Data are from Kaiser Permanente Northern California, an integrated health care delivery system. This study used a volunteer sample who completed the surveys. Participants included Chinese, Filipino, and Japanese individuals who were aged 60 to less than 90 years without a dementia diagnosis in the EHR at the time of the survey (baseline) and who had 2 years of health plan coverage before baseline. Data analysis was performed from December 2021 to December 2022. Exposures: The main exposure was educational attainment (college degree or higher vs less than a college degree), and the main stratification variables were Asian ethnicity and nativity (born in the US or born outside the US). Main Outcomes and Measures: The primary outcome was incident dementia diagnosis in the EHR. Dementia incidence rates were estimated by ethnicity and nativity, and Cox proportional hazards and Aalen additive hazards models were fitted for the association of college degree or higher vs less than a college degree with time to dementia, adjusting for age (timescale), sex, nativity, and an interaction between nativity and college degree. Results: Among 14 749 individuals, the mean (SD) age at baseline was 70.6 (7.3) years, 8174 (55.4%) were female, and 6931 (47.0%) had attained a college degree. Overall, among individuals born in the US, those with a college degree had 12% lower dementia incidence (HR, 0.88; 95% CI, 0.75-1.03) compared with those without at least a college degree, although the confidence interval included the null. The HR for individuals born outside the US was 0.82 (95% CI, 0.72-0.92; P = .46 for the college degree by nativity interaction). The findings were similar across ethnicity and nativity groups except for Japanese individuals born outside the US. Conclusions and Relevance: These findings suggest that college degree attainment was associated with lower dementia incidence, with similar associations across nativity. More work is needed to understand determinants of dementia in Asian American individuals and to elucidate mechanisms linking educational attainment and dementia.

Prevalence and changes of low-value care at acute care hospitals: a multicentre observational study in Japan.

OBJECTIVES: We aimed to examine the use and factors associated with the provision of low-value care in Japan. DESIGN: A multicentre observational study. SETTING: Routinely collected claims data that include all inpatient and outpatient visits in 242 large acute care hospitals (accounting for approximately 11% of all acute hospitalisations in Japan). PARTICIPANTS: 345 564 patients (median age (IQR): 62 (40-75) years; 182 938 (52.9%) women) seeking care at least once in the hospitals in the fiscal year 2019. PRIMARY AND SECONDARY OUTCOME MEASURES: We identified 33 low-value services, as defined by clinical evidence, and developed two versions of claims-based measures of low-value services with different sensitivity and specificity (broader and narrower definitions). We examined the number of low-value services, the proportion of patients receiving these services and the proportion of total healthcare spending incurred by these services in 2019. We also evaluated the 2015-2019 trends in the number of low-value services. RESULTS: Services identified by broader low-value care definition occurred in 7.5% of patients and accounted for 0.5% of overall annual healthcare spending. Services identified by narrower low-value care definition occurred in 4.9% of patients and constituted 0.2% of overall annual healthcare spending. Overall, there was no clear trend in the prevalence of low-value services between 2015 and 2019. When focusing on each of the 17 services accounting for more than 99% of all low-value services identified (narrower definition), 6 showed decreasing trends from 2015 to 2019, while 4 showed increasing trends. Hospital size and patients' age, sex and comorbidities were associated with the probability of receiving low-value service. CONCLUSIONS: A substantial number of patients received low-value care in Japan. Several low-value services with high frequency, especially with increasing trends, require further investigation and policy interventions for better resource allocation.

Prognosis of high-risk human papillomavirus-related cervical lesions: A hidden Markov model analysis of a single-center cohort in Japan.

INTRODUCTION: Previous studies have shown that individuals with human papillomavirus (HPV)-related cervical lesions have different prognoses according to the HPV genotype. However, these studies failed to account for possible diagnostic misclassification. In this retrospective cohort study, we aimed to clarify the natural course of cervical lesions according to HPV genotype to account for any diagnostic misclassification. MATERIALS AND METHODS: Our cohort included 729 patients classified as having cervical intraepithelial neoplasia (CIN). HPV was genotyped in all patients, who were followed up or treated for cervical lesions at the University of Tokyo Hospital from October 1, 2008 to March 31, 2015. Hidden Markov models were applied to estimate the diagnostic misclassification probabilities of the current diagnostic practice (histology and cytology) and the transitions between true states. We then simulated two-year transition probabilities between true cervical states according to HPV genotype. RESULTS: Compared with lesions in patients with other HPV genotypes, lesions in HPV 16-positive patients were estimated to be more likely to increase in severity (i.e., CIN3/cancer); over 2 years, 17.7% (95% confidence interval [CI], 9.3%-29.3%) and 27.8% (95% CI, 16.6%-43.5%) of those with HPV 16 progressed to CIN3/cancer from the true states of CIN1 and CIN2, respectively, whereas 55%-70% of CIN1/2 patients infected with HPV 52/58 remained in the CIN1/2 category. Misclassification was estimated to occur at a rate of 3%-38% in the current diagnostic practice. CONCLUSION: This study contributes robust evidence to current literature on cervical lesion prognosis according to HPV genotype and quantifies the diagnostic misclassification of true cervical lesions.

Claims-based algorithms for common chronic conditions were efficiently constructed using machine learning methods.

Identification of medical conditions using claims data is generally conducted with algorithms based on subject-matter knowledge. However, these claims-based algorithms (CBAs) are highly dependent on the knowledge level and not necessarily optimized for target conditions. We investigated whether machine learning methods can supplement researchers' knowledge of target conditions in building CBAs. Retrospective cohort study using a claims database combined with annual health check-up results of employees' health insurance programs for fiscal year 2016-17 in Japan (study population for hypertension, N = 631,289; diabetes, N = 152,368; dyslipidemia, N = 614,434). We constructed CBAs with logistic regression, k-nearest neighbor, support vector machine, penalized logistic regression, tree-based model, and neural network for identifying patients with three common chronic conditions: hypertension, diabetes, and dyslipidemia. We then compared their association measures using a completely hold-out test set (25% of the study population). Among the test cohorts of 157,822, 38,092, and 153,608 enrollees for hypertension, diabetes, and dyslipidemia, 25.4%, 8.4%, and 38.7% of them had a diagnosis of the corresponding condition. The areas under the receiver operating characteristic curve (AUCs) of the logistic regression with/without subject-matter knowledge about the target condition were .923/.921 for hypertension, .957/.938 for diabetes, and .739/.747 for dyslipidemia. The logistic lasso, logistic elastic-net, and tree-based methods yielded AUCs comparable to those of the logistic regression with subject-matter knowledge: .923-.931 for hypertension; .958-.966 for diabetes; .747-.773 for dyslipidemia. We found that machine learning methods can attain AUCs comparable to the conventional knowledge-based method in building CBAs.

Association of managerial position with cardiovascular risk factors: A fixed-effects analysis for Japanese employees.

OBJECTIVES: Although higher occupational classes have been reported to be associated with better health, researchers do not fully understand whether such associations derive from the position or individual characteristics of the person in that position. We examined the association between being a manager and cardiovascular disease (CVD) risk factors using unique panel data in Japan that annually observed employees' occupational class and health conditions. METHODS: We analyzed data for 45 888 observations from a Japanese company from 2013 through 2017. The association between being a manager and CVD risk factors (metabolic risks and health-related behaviors) were evaluated using simple pooled cross-sectional analyses with adjustment for age, sex, marital status, and overtime-working hours. We further incorporated employee-level fixed-effects into the models to examine whether the associations were subject to individual time-invariant factors. RESULTS: The pooled cross-sectional analyses showed that, compared to non-managers, managers had 2.0 mg/dl lower low density lipoprotein cholesterol (LDL-C) level, 1.4 mmHg-lower systolic blood pressure, and 0.2 kg/m2 lower body mass index (BMI). After adjusting for employee-level fixed-effects, being a manager was associated with a significantly 2.2 mg/dl higher LDL-C level. However, the associations between an individual's management status and blood pressure or BMI were not significant. Furthermore, managers were 5.5% less likely to exercise regularly and 6.1% less likely to report sufficient sleep in the fixed-effects models, although the pooled cross-sectional analyses did not demonstrate these significant associations. CONCLUSIONS: Our findings suggest the necessity of considering these unfavorable health risks associated with being promoted to a manager.

Physician Distribution by Specialty and Practice Setting: Findings in Japan in 2000, 2010 and 2016.

As the medical demand is projected to increase along with the population aging in Japan, the geographical distribution of physicians is a significant concern for society and policymakers. To implement effective measures on geographical physician distribution, this study aimed to describe and compare the distribution of physicians by specialty in 2000, 2010 and 2016 in Japan, and examine whether practice setting was associated with distribution. To quantify the geographical physician distribution by specialty, we calculated the Gini coefficients of physicians working at clinics or hospitals in 2000, 2010, and 2016. We used the basic geographic unit for medical care planning in Japan, a secondary medical area, as the study unit. To show the association between the geographical distribution of physicians in each specialty and their practice setting, we categorized specialties into two groups by the proportion of physicians in that specialty working in hospitals, and showed aggregated Lorenz curves for each category. The overall geographical distribution of physicians appeared to improve during the study period, but varied by specialty. Those in specialties, where at least 90% of physicians work in hospitals such as anesthesiologists and radiologists, were more clustered, as shown by the Lorenz curves and the Gini coefficients. Similar distributional differences were also found even when we excluded physicians working in clinics, meaning that the distributional variation could be explained by other factors than the distribution of hospitals. These results suggest that the nature of practice in each specialty strongly affects the geographical distribution of specialists.