Patient-Physician Sex Discordance and “Before Medically Advised” Discharge from Hospital: A Population-Based Retrospective Cohort Study

Abstract

Background

Patient-physician sex discordance (when patient sex does not match physician sex) has been associated with reduced clinical rapport and adverse outcomes including post-operative mortality and unplanned hospital readmission. It remains unknown whether patient-physician sex discordance is associated with “before medically advised” hospital discharge (BMA discharge; commonly known as discharge “against medical advice”).

Objective

To evaluate whether patient-physician sex discordance is associated with BMA discharge.

Design

Retrospective cohort study using 15 years (2002–2017) of linked population-based administrative health data for all non-elective, non-obstetrical acute care hospitalizations from British Columbia, Canada.

Participants

All individuals with eligible hospitalizations during study interval.

Main Measures

Exposure: patient-physician sex discordance. Outcomes: BMA discharge (primary), 30-day hospital readmission or death (secondary).

Results

We identified 1,926,118 eligible index hospitalizations, 2.6% of which ended in BMA discharge. Among male patients, sex discordance was associated with BMA discharge (crude rate, 4.0% vs 2.9%; adjusted odds ratio [aOR] 1.08; 95%CI 1.03–1.14; p = 0.003). Among female patients, sex discordance was not associated with BMA discharge (crude rate, 2.0% vs 2.3%; aOR 1.02; 95%CI 0.96–1.08; p = 0.557). Compared to patient-physician sex discordance, younger patient age, prior substance use, and prior BMA discharge all had stronger associations with BMA discharge.

Conclusions

Patient-physician sex discordance was associated with a small increase in BMA discharge among male patients. This finding may reflect communication gaps, differences in the care provided by male and female physicians, discriminatory attitudes among male patients, or residual confounding. Improved communication and better treatment of pain and opioid withdrawal may reduce BMA discharge.

Supplementary Information

The online version contains supplementary material available at 10.1007/s11606-024-08697-8.

INTRODUCTION

Every year, about 500,000 hospitalizations in the United States end in departure from hospital before completion of medical treatment^1,2. Such departures are described as patient-initiated, patient-directed, “against medical advice,” or before medically advised (BMA) discharge from hospital³. Individuals departing hospital without medical clearance are up to seven times more likely to have an unplanned readmission within 2 weeks and up to three-fold more likely to die within a year^4–7. Patient characteristics commonly associated with BMA discharge include male sex, younger age, socioeconomic marginalization, housing instability, presence of specific medical diagnoses (substance use disorder, mental illness, human immunodeficiency virus [HIV] infection, cirrhosis, asthma), and a personal history of prior BMA discharge⁸.

Patient-physician sex discordance (when the sex of the patient differs from the sex of the treating physician) is a plausible but previously unstudied risk factor for BMA discharge^9–13. A systematic review of 10 eligible studies found that patient-physician sex discordance influenced consultation length, patient openness, use of nonverbal communication, demonstrations of power and status, and patient-centeredness¹³. Sex-concordant dyads displayed more calmness, ease, and equality in discussions. Sex-discordant dyads with female physicians exhibited the least friendliness and calmness, with more conflicts around gender roles, while sex-discordant dyads with male physicians were the least patient-centered and more presumptuous about patients’ histories¹³. Moreover, male and female patients may prefer distinct communication styles, as was illustrated in a lab experiment that simulated medical visits with computer-generated virtual physicians: female patients reported greater satisfaction with “caring” female physicians and with “caring” but “non-dominant” male physicians, but satisfaction among male patients was not associated with physician communication style (caring or dominance)¹¹. Patient-physician sex discordance has also been associated with adverse patient outcomes. Among 1.3 million elective surgery patients, patient-surgeon sex discordance was linked to worse post-operative outcomes in female patient-male surgeon dyads, but not among male patient-female surgeon dyads¹⁴. Another study found that female patients treated by male physicians were more likely to die after myocardial infarction than male patients treated by female physicians¹⁰. Patient-physician sex discordance thus appears to affect patient-physician rapport and modify clinical communication, with consequences for patient outcomes.

In light of these findings, we hypothesized that patient-physician sex discordance might increase the risk of BMA discharge. We examined this possibility by performing a retrospective observational cohort study with 15 years of population-based health administrative data.

METHODS

Data Sources

This study was based on linked and de-identified population-based administrative health data from British Columbia (BC), Canada (Supplementary Appendix, eMethod 1). Our data included the following: patient demographics from the Consolidation and Income Band files; hospital admission records from the Discharge Abstract Database (DAD); physician services data from the Medical Services Plan (MSP) fee-for-service payment file; physician characteristics from the MSP Practitioner File; data on all outpatient prescription medications dispensed by community pharmacies in BC from PharmaNet; and death records from Vital Statistics (eMethod 1).

Study Cohort, Exposure, and Outcome

The study cohort included all urgent (non-elective) admissions to any acute care hospital in BC with a discharge date between 1 January 2002 and 31 March 2017. We were primarily interested in non-elective hospitalizations because elective hospitalizations are typically brief, occur mainly to facilitate elective procedures, and rarely end in BMA discharge (Supplementary Appendix, eTable 1)¹⁵. We excluded hospitalizations when the discharge diagnosis corresponded to pregnancy, childbirth, or the puerperal/perinatal periods (because these rarely result in BMA discharge); patient sex was missing; patient age was < 18 years (because minors are typically not allowed to depart hospital without parental consent); the admission was for an elective same-day procedure; it began or ended with a transfer from or to another acute care hospital (because it was not possible to determine which hospital setting influenced their treatment); it ended in discharge to long-term care, palliative care, a rehabilitation or pediatric hospital, or an addiction treatment center (because these patients are typically too frail, medically unwell, or dependent on nursing care to be at substantial risk of BMA discharge); and when the hospitalization could not be linked with characteristics of the primary treating physician.

Our primary exposure was patient-physician sex discordance. Patient sex was obtained from BC Vital Statistics. We identified the primary treating physician using the index hospitalization’s Most Responsible Physician (defined as “the provider who was most responsible for the patient’s care during the hospitalization”; eTable 1)¹⁶. The Most Responsible Physician is normally the clinician responsible for the greatest proportion of the length of stay, and is established by professional medical record abstractors after detailed review of hospital records. Physician sex was obtained from the College of Physicians and Surgeons of British Columbia and is established via an identity document at the time of professional registration. These variables reflect binary classification of biological sex (“male,” “female”). We performed separate analyses in male and female patients to better understand how the effect of patient-physician sex discordance differed by patient sex. We considered sex discordance to be present when a hospitalized patient was recorded to have a Most Responsible Physician of the opposite sex. Our primary outcome was BMA discharge, an explicitly coded disposition category within the DAD. As a secondary outcome, we evaluated a composite of unplanned hospital readmission or death within 30 days of index hospitalization discharge.

Analyses

We used a multilevel logistic regression model to evaluate the influence of patient-physician sex discordance on BMA discharge while accounting for potential patient-, physician-, and hospital-level confounders. Patient characteristics, features of the index hospitalization, and physician characteristics were modeled as fixed effects at level 1. The random intercepts at levels 2 and 3 were anonymized physician and hospital site identifiers, respectively. Model fit was assessed using visual examination of the standardized residuals. Individuals with multiple eligible hospital admissions contributed all eligible hospitalizations to the analysis at level 1. Including only one hospitalization per person would underrepresent individuals with multiple BMA discharges and yield biased results¹⁷.

We adjusted the multivariable models for potential confounders identified through literature review and clinical experience (eTable 1). We excluded covariates with strong collinearity. Patient characteristics included age; household income quintile and population density of patient’s residential neighborhood (rural vs urban vs missing); features of the index hospitalization (e.g., length of stay, discharge diagnosis, surgical or therapeutic interventions performed [yes/no], intensive care unit [ICU] stay, study month of index discharge date [a sequential number assigned to the index visit month to account for period effects and the increase in female physicians over time]); health service use in the year prior to index hospitalization (e.g., hospitalizations, clinic visits); Charlson Comorbidity Index ≥ 2 and substance use comorbidities (deemed present if identified in diagnostic coding associated with ≥ 1 hospitalization or ≥ 2 physician visits in the year prior to index admission); and prior BMA discharge. In addition to assessing the influence of physician sex, we included other physician characteristics associated with patient outcomes: years of clinical experience (years between medical school graduation and index hospitalization year) and location of graduating medical school (British Columbia vs elsewhere in Canada vs international)^12,18–20 We performed exploratory subgroup analyses based on level 1 variables.

Ethics

The University of British Columbia Clinical Research Ethics Board approved the study and waived the requirement for individual consent (H17-01039). Data were de-identified before release to investigators. All inferences, opinions, and conclusions drawn in this publication are those of the authors and do not reflect the opinions or policies of the Data Stewards.

RESULTS

Our final sample included 1,926,118 eligible index hospitalizations involving 937,021 unique patients and 8843 unique physicians (Fig. 1; Supplementary Appendix, eFigures 1 and 2). About half of the hospitalizations involved a female patient yet only 20% involved a female physician. In total, 49,802 (2.6%) of all hospitalizations ended in BMA discharge.

Figure 1.

Flow diagram describing cohort selection.

Male patients were far less likely than female patients to be exposed to patient-physician sex discordance (18% vs 77%, p < 0.001) but were more likely to choose BMA discharge (3.1% vs 2.0%, p < 0.001). Male patients were also more likely to have prior BMA discharges and prior alcohol and drug use (eTable 2). Relative to sex-concordant male dyads, male patients with a female physician were more likely to be admitted to Medicine or Psychiatry, more likely to have Charlson Comorbidity Index ≥ 2, and more likely to have substance use listed as the Most Responsible Diagnosis (Table 1). In contrast, relative to sex-concordant female dyads, female patients with a male physician were more likely to be admitted to Surgery, less likely to have Charlson Comorbidity Index ≥ 2, and less likely to have substance use listed as the Most Responsible Diagnosis (Table 2). Prior BMA discharge was nine times more common among index hospitalizations ending in BMA discharge relative to index hospitalizations ending with physician approval (Table 3).

Table 1.

Characteristics Associated with Index Hospitalization Among Male Patients

Characteristics	Sex-concordant dyads (male patient-male physician) N = 800,295 (%)	Sex-discordant dyads (male patient-female physician) N = 178,652 (%)	SMD
Patient characteristics
Demographics
Median age (years) [Q1, Q3]	62 [46, 75]	61 [45, 76]	0.014*
Neighborhood income quintile			0.048*
First (lowest)	203,485 (25.4)	48,606 (27.2)
Second	162,652 (20.3)	35,043 (19.6)
Third	146,513 (18.3)	32,316 (18.1)
Fourth	137,581 (17.2)	30,580 (17.1)
Fifth (highest)	129,036 (16.1)	27,022 (15.1)
Urban residence	224,321 (28.0)	57,021 (31.9)	0.085*
Medical history
≥ 1 hospitalizations in the past year	384,280 (48.0)	90,256 (50.5)	0.050*
≥ 1 BMA discharge in the past year	25,472 (3.2)	8276 (4.6)	0.075*
≥ 7 physician clinic visits in past year	136,568 (17.1)	30,055 (16.8)	0.006+
Charlson Comorbidity Index ≥ 2	158,425 (19.8)	38,432 (21.5)	0.042*
Comorbidity
Opioid use disorder	9431 (1.2)	3656 (2.0)	0.069*
Alcohol use disorder	39,984 (5.0)	11,850 (6.6)	0.070*
Other substance use disorder	35,796 (4.5)	12,522 (7.0)	0.109*
Psychiatric disorders	85,844 (10.7)	25,502 (14.3)	0.107*
Diabetes (with or without complications)	69,818 (8.7)	18,804 (10.5)	0.061*
Myocardial infarction	31,549 (3.9)	6241 (3.5)	0.024*
Cerebrovascular disease	18,745 (2.3)	4205 (2.4)	0.001
Chronic obstructive pulmonary disease	53,776 (6.7)	12,965 (7.3)	0.021*
Renal disease	31,849 (4.0)	8951 (5.0)	0.050*
Cancer	58,960 (7.4)	12,543 (7.0)	0.013*
Dementia	8893 (1.1)	2542 (1.4)	0.028*
HIV	3891 (0.5)	1705 (1.0)	0.055*
Medications
≥ 2 prescription medications at baseline	497,031 (62.1)	114,636 (64.2)	0.269*
Active prescription medications (90 days prior to hospital admission) – median [Q1, Q3]	3 [0, 6]	3 [1, 7]	0.226*
Selected prescription medications filled in the 90 days prior to index date
Opioids	181,308 (22.7)	40,629 (22.7)	0.166
OAT	1725 (0.2)	678 (0.4)	0.073*
Benzodiazepine	121,488 (15.2)	29,118 (16.3)	0.213*
Beta blockers	159,162 (19.9)	37,329 (20.9)	0.241*
Antipsychotics	77,194 (9.6)	23,355 (13.1)	0.254*
Antidepressants	136,511 (17.1)	35,918 (20.1)	0.177*
Details of index hospitalization
Median length of stay (days) [Q1, Q3]	4 [2, 7]	4 [2, 8]	0.053*
ICU stay	82,387 (10.3)	14,719 (8.2)	0.071*
Most Responsible Diagnosis			0.206*
Psychiatric	61,771 (7.7)	19,873 (11.1)
Cardiac disease	30,944 (3.9)	4840 (2.7)
Obstructive lung disease	34,963 (4.4)	9369 (5.2)
Substance use	36,761 (4.6)	11,661 (6.5)
Pneumonia	26,599 (3.3)	7319 (4.1)
Biliary tract disease	28,419 (3.6)	6027 (3.4)
Most Responsible Service			0.328*
Medicine	531,722 (66.4)	131,816 (73.8)
Psychiatry	75,135 (9.4)	24,097 (13.5)
Surgery	188,168 (23.5)	20,807 (11.6)
Other	5270 (0.7)	1931 (1.1)
Physician characteristics
Median age (years) [Q1, Q3]	49 [42, 57]	43 [37, 50]	0.609*
Median years of clinical experience [Q1, Q3]	23 [15, 31]	15 [9, 23]	0.668*
Location of medical school			0.208*
British Columbia	203,076 (25.4)	58,367 (32.7)
Other Canadian medical school	356,281 (44.5)	80,730 (45.2)
International medical school	240,938 (30.1)	39,555 (22.1)
Specialty			0.449*
Primary care	372,119 (46.5)	88,864 (49.7)
Internal medicine	121,271 (15.2)	34,148 (19.1)
Psychiatry	58,427 (7.3)	20,339 (11.4)
General surgery	70,790 (8.8)	11,814 (6.6)
Other medical	62,018 (7.7)	15,457 (8.7)
Other surgical	106,254 (13.3)	4330 (2.4)
Other	9416 (1.2)	3700 (2.1)
Hospital size			0.103*
Large	256,695 (32.1)	49,789 (27.9)
Medium	141,493 (17.7)	33,939 (19.0)
Small	65,471 (8.2)	17,752 (9.9)
Teaching	333,538 (41.7)	76,329 (42.7)
Missing	3098 (0.4)	843 (0.5)

Table comparing characteristics among male patients treated by male vs female physicians. Each diagnosis in the Charlson Comorbidity Index was based on comorbidities with ≥ 1 hospitalization (DAD) or ≥ 2 clinic visits (MSP) in the year prior to index hospital admission. Active medication at baseline denotes a prescription fill on the date of the index hospital admission. The category Other in Most Responsible Service includes primary care practice, Gynecology, Pediatrics, Anesthesia, Alternate Level of Care, and missing. The variable “Specialty” in physician characteristics denotes the specialty of the Most Responsible Physician; the category Other includes Gynecology, Pediatrics, Neurology, Radiology, Pathology, Emergency Medicine, Chiropractor, Public Health, Nuclear Medicine, Naturopathy, Physical Therapy, Medical Genetics, and missing. In Canada, primary care physicians are trained in “family medicine” or “general practice”; internal medicine and its subspecialties only provide consultant care and are not trained to provide primary care. SMD indicates standardized mean difference, * denotes p < 0.001, ⁺ denotes p < 0.05. OAT, opioid agonist therapy for opioid use disorder (including methadone, buprenorphine-naloxone, and long-acting oral morphine)

Table 2.

Characteristics of Index Hospitalization Among Female Patients

Characteristics	Sex-concordant dyads (female patient-female physician) N = 214,805 (%)	Sex-discordant dyads (female patient-male physician) N = 732,366 (%)	SMD
Patient characteristics
Demographics
Median age (years) [Q1, Q3]	63 [44, 79]	63 [46, 78]	0.017*
Neighborhood income quintile			0.027*
First (lowest)	55,944 (26.0)	194,323 (26.5)
Second	43,710 (20.3)	153,905 (21.0)
Third	39,572 (18.4)	134,327 (18.3)
Fourth	37,565 (17.5)	122,725 (16.8)
Fifth (highest)	33,116 (15.4)	111,143 (15.2)
Urban residence	65,301 (30.4)	207,208 (28.3)	0.047*
Medical history
≥ 1 hospitalizations in the past year	107,445 (50.0)	354,170 (48.4)	0.033*
≥ 1 BMA discharge in the past year	5699 (2.7)	16,372 (2.2)	0.027*
≥ 7 physician clinic visits in past year	42,134 (19.6)	146,302 (20.0)	0.009*
Charlson Comorbidity Index ≥ 2	38,047 (17.7)	118,243 (16.1)	0.042*
Comorbidity
Opioid use disorder	3231 (1.5)	7887 (1.1)	0.038*
Alcohol use disorder	7041 (3.3)	19,858 (2.7)	0.033*
Other substance use disorder	8814 (4.1)	23,822 (3.3)	0.045*
Psychiatric disorders	32,323 (15.0)	96,587 (13.2)	0.053*
Diabetes (with or without complications)	17,482 (8.1)	52,344 (7.1)	0.037*
Myocardial infarction	4565 (2.1)	17,210 (2.3)	0.015*
Cerebrovascular disease	4366 (2.0)	15,408 (2.1)	0.005+
Renal disease	8612 (4.0)	25,082 (3.4)	0.031*
Cancer	13,713 (6.4)	41,776 (5.7)	0.029*
Dementia	3140 (1.5)	9159 (1.3)	0.018*
HIV	706 (0.3)	1709 (0.2)	0.018*
Medications
≥ 2 prescription medications at baseline	150,785 (70.2)	509,661 (69.6)	0.328*
Active prescription medications (90 days prior to hospital admission) – median [Q1, Q3]	4 [1, 7]	3 [1, 7]	0.231*
Selected prescription medications filled in the 90 days prior to index date
Opioids	56,400 (26.3)	196,948 (26.9)	0.121*
OAT	631 (0.3)	1508 (0.2)	0.075*
Benzodiazepine	48,849 (22.7)	167,561 (22.9)	0.242
Beta-blockers	39,954 (18.6)	133,244 (18.2)	0.300*
Antipsychotics	27,559 (12.8)	82,124 (11.2)	0.312*
Antidepressants	62,588 (29.1)	202,111 (27.6)	0.182*
Details of index hospitalization
Median length of stay (days) [Q1, Q3]	4 [2, 8]	4 [2, 8]	0.023*
ICU stay	11,779 (5.5)	49,964 (6.8)	0.056*
Most Responsible Diagnosis			0.117*
Psychiatric	23,465 (10.9)	64,593 (8.8)
Cardiac disease	19,556 (9.1)	74,261 (10.1)
Obstructive lung disease	11,033 (5.1)	34,307 (4.7)
Substance use	8204 (3.8)	24,052 (3.3)
Pneumonia	7965 (3.7)	23,789 (3.2)
Biliary tract disease	7686 (3.6)	33,454 (4.6)
Most Responsible Service			0.304*
Medicine	153,157 (71.3)	492,171 (67.2)
Psychiatry	25,144 (11.7)	69,416 (9.5)
Surgery	24,407 (11.4)	153,419 (20.9)
Other	12,096 (5.6)	17,360 (2.4)
Physician characteristics
Median age (years) [Q1, Q3]]	44 [37, 51]	50 [42, 57]	0.548*
Median years of clinical experience [Q1, Q3]	16 [9, 24]	23 [15, 31]	0.599*
Location of medical school			0.246*
British Columbia	72,102 (33.6)	178,812 (24.4)
Other Canadian medical school	93,859 (43.7)	319,868 (43.7)
International medical school	48,844 (22.7)	233,686 (31.9)
Specialty			0.434*
Primary care	125,732 (58.5)	380,873 (52.0)
Internal medicine	27,482 (12.8)	91,045 (12.4)
Psychiatry	20,100 (9.4)	54,064 (7.4)
General surgery	12,479 (5.8)	71,122 (9.7)
Other medical	11,958 (5.6)	44,290 (6.0)
Other surgical	3230 (1.5)	70,584 (9.6)
Other	13,824 (6.4)	20,388 (2.8)
Hospital size			0.128*
Large	64,224 (29.9)	248,027 (33.9)
Medium	47,395 (22.1)	135,285 (18.5)
Small	23,724 (11.0)	65,895 (9.0)
Teaching	78,911 (36.7)	281,010 (38.4)
Missing	551 (0.3)	2149 (0.3)

Table comparing characteristics among female patients treated by female vs male physicians. Select variable definitions are in Table 1 legend; full list is in eTable 1. SMD indicates standardized mean difference, * denotes p < 0.001, ⁺ denotes p < 0.05. OAT, opioid agonist therapy for opioid use disorder (including methadone, buprenorphine-naloxone, and long-acting oral morphine)

Table 3.

Characteristics Associated with Index Hospitalizations Ending in BMA vs Physician-Advised Discharge

Characteristics	BMA discharge, count (%) n = 49,802	PA discharge, count (%) n = 1,876,316	SMD
Patient characteristics
Demographics
Male sex	30,561 (61.4)	948,386 (50.5)	0.219*
Median age (years) [Q1, Q3]	46 [34, 58]	63 [46, 77]	0.734*
Age groups			0.625*
≤ 29 years	8033 (16.1)	167,035 (8.9)
30–49 years	21,367 (42.9)	386,098 (20.6)
≥ 50 years	20,402 (41.0)	1,323,183 (70.5)
Neighborhood income quintile			0.396*
1 (lowest)	19,753 (39.7)	482,605 (25.7)
2	9504 (19.1)	385,806 (20.6)
3	7103 (14.3)	345,625 (18.4)
4	5679 (11.4)	322,772 (17.2)
5 (highest)	4896 (9.8)	295,421 (15.7)
Missing	2867 (5.8)	44,087 (2.3)
Urban residence	14,440 (29.0)	539,411 (28.7)	0.028*
Medical history
≥ 1 hospitalizations in the past year	28,451 (57.1)	907,700 (48.4)	0.176*
≥ 1 hospitalizations ending in BMA discharge in the past year	11,061 (22.2)	44,758 (2.4)	0.633*
≥ 7 physician clinic visits in the past year	8355 (16.8)	346,704 (18.5)	0.045*
≥ 1 psychiatric hospitalization in the past year	9710 (19.5)	115,659 (6.2)	0.407*
Charlson Comorbidity Score ≥ 2	8190 (16.4)	344,957 (18.4)	0.051*
Comorbidity
Opioid use disorder	4681 (9.4)	19,524 (1.0)	0.383*
Alcohol use disorder	8355 (16.8)	70,378 (3.8)	0.439*
Other substance use disorder	11,504 (23.1)	69,450 (3.7)	0.594*
Psychiatric disorders	12,192 (24.5)	228,064 (12.2)	0.323*
Diabetes (with or without complications)	3801 (7.6)	154,647 (8.2)	0.047*
Myocardial infarction	790 (1.6)	58,775 (3.1)	0.102*
Congestive heart failure	1822 (3.7)	120,091 (6.4)	0.126*
Cerebrovascular disease	726 (1.5)	41,998 (2.2)	0.058*
Chronic obstructive pulmonary disease	3285 (6.6)	129,947 (6.9)	0.013+
Renal disease	1422 (2.9)	73,072 (3.9)	0.058*
Cancer	1434 (2.9)	125,558 (6.7)	0.179*
Dementia	472 (0.9)	23,262 (1.2)	0.028*
HIV	1723 (3.5)	62,88 (0.3)	0.231*
Liver disease	3156 (6.3)	35,708 (1.9)	0.224*
Details of index hospitalization
Median length of stay (days) [Q1, Q3]	3 [1, 6]	4 [2, 8]	0.104*
Patient-physician sex-discordant dyads	21,545 (43.3)	889,473 (47.4)	0.083*
Most Responsible Diagnosis			0.629*
Psychiatric	7296 (14.7)	162,406 (8.7)
Cardiac disease	3133 (6.3)	224,607 (12.0)
Substance use	7887 (15.8)	72,791 (3.9)
Biliary tract disease	1404 (2.8)	74,182 (4.0)
Cancer	608 (1.2)	66,350 (3.5)
Diabetes	1518 (3.0)	31,276 (1.7)
Cellulitis	2061 (4.1)	22,500 (1.2)
HIV	1135 (2.3)	3950 (0.2)
Most Responsible Service for index hospitalization			0.481*
Medicine	35,006 (70.3)	1,273,860 (67.9)
Psychiatry	10,657 (21.4)	183,135 (9.8)
Surgery	3583 (7.2)	383,218 (20.4)
Physician characteristics
Median age (years) [Q1, Q3]	47 [40, 55]	48 [40, 56]	0.064*
Median years of clinical experience [Q1, Q3]	21 [12, 29]	21 [13, 30]	0.059*
Location of medical school			0.090*
British Columbia	11,635 (23.4)	500,722 (26.7)
Other Canadian medical school	21,989 (44.2)	828,749 (44.2)
International medical school	16,178 (32.5)	546,845 (29.1)
Specialty			0.415*
Primary care	25,086 (50.4)	942,502 (50.2)
Internal medicine	9360 (18.8)	264,586 (14.1)
Psychiatry	7082 (14.2)	145,848 (7.8)
General surgery	1609 (3.2)	164,596 (8.8)
Patient-physician dyads			0.245*
Female patient-female physician	4893 (9.8)	209,912 (11.2)
Female patient-male physician	14,348 (28.8)	718,018 (38.3)
Male patient-female physician	7197 (14.5)	171,455 (9.1)
Male patient-male physician	23,364 (46.9)	776,931 (41.4)
Hospital size			0.199*
Large	12,611 (25.3)	606,124 (32.3)
Medium	8130 (16.3)	349,982 (18.7)
Small	4527 (9.1)	168,315 (9.0)
Teaching	24,383 (49.0)	745,405 (39.7)

Comparison of patient, physician, and hospital characteristics between hospitalizations ending in before medically advised (BMA) discharge vs physician-advised (PA) discharge. Select variable definitions are in Table 1 legend; full definition list is in eTable 1. SMD indicates standardized mean difference, * denotes p < 0.001, ⁺ denotes p < 0.05

In our primary analysis, patient-physician sex discordance was associated with a small increase in the risk of BMA discharge among male patients (n = 978,947; crude rate, 4.0% vs 2.9%; adjusted odds ratio [aOR], 1.08; 95%CI 1.03–1.14; p < 0.003; Fig. 2, eTable 3). We found no such association among female patients (n = 947,171; crude rate, 2.0% vs 2.3%; aOR, 1.02; 95%CI 0.96–1.08; p = 0.557). Other patient and physician characteristics were also associated with BMA discharge (eTable 3). An additional decade of physician clinical experience was associated with about a 25% reduction in the odds of BMA discharge. Compared with physicians who graduated from a medical school in British Columbia, physicians graduating from international medical schools were associated with a small increase in the odds of BMA discharge. Relative to patient-physician sex discordance, a number of patient characteristics that exhibited a relatively strong association with BMA discharge included age 30–49 years (two-fold increase in risk), a history of drug use (two-fold increase in risk), a history of prior BMA discharge (four-fold increase in risk), and a Most Responsible Diagnosis of cellulitis, HIV, or substance use⁸. Variables indicating higher acuity (i.e., arrival by ambulance, admission via emergency department, ICU stay) were also associated with increased risk of BMA discharge.

Figure 2.

Effect of patient-physician sex discordance on BMA discharge among key subgroups

Results of subgroup analyses were consistent with the primary analysis, with the strongest effect being among male surgery patients (aHR, 1.31; Fig. 2, eTable 4). The association between sex discordance and BMA discharge was relatively consistent over the 15-year study interval (eFigure 3). Among male patients, patient-physician sex discordance was associated with an increased risk of secondary outcomes including readmission or death and death alone within 30 days of index discharge (Fig. 2; eTable 4). In contrast, among female patients, sex discordance was associated with a reduced risk of either readmission or death and death alone within 30 days of discharge.

DISCUSSION

In this population-based retrospective cohort study of 1.9 million hospitalizations over a 15-year study interval, we found that patient-physician sex discordance was associated with a small increase in BMA discharge among male patients, while a similar increase was not observed among female patients. Similarly, patient-physician sex discordance was also associated with an increased risk of unplanned readmission or death within 30 days of index discharge among male patients but not among female patients. Relative to patient-physician sex discordance, patient characteristics such as male sex, younger age, prior substance use, and prior BMA discharge were stronger risk factors for BMA discharge.

One interpretation of our results suggests that patient-physician sex discordance might contribute in subtle ways to failures of communication between male patients and female physicians, arising from differences in values, nonverbal behaviors, and preferred styles of communication^{9,11, 13, 21, 22}. This may erode trust, encourage treatment non-adherence, and provoke adverse outcomes like BMA discharge, unplanned readmission, and death²³. Arguing against this explanation are studies suggesting female physicians have fewer disagreements with patients²⁴, better patient rapport, and higher patient satisfaction scores^25,26, all of which might be expected to reduce the risk of BMA discharge. Perhaps caring and empathetic communication styles may be valued more highly by female patients, but may not substantially influence decision-making among male patients who initiate a BMA discharge¹¹.

Another interpretation focuses on differences in the medical care provided by male and female physicians. Prior research suggests female physicians perform more investigations, emphasize preventative services, and attend to psychosocial issues, potentially prolonging length of stay and increasing the likelihood of BMA discharge among more risk-tolerant male patients^{9,13, 27–30}. In the primary care setting, this might explain why female physicians’ patients exhibit a lower risk of readmission and death^12,31. In a hospital setting, this predisposition toward patient-centered care suggests female physicians are more likely to partner with patients to provide harm-reducing discharge prescriptions and follow-up appointments when a BMA discharge cannot be prevented. However, we also observed an increase in unplanned readmission and death among male patient-female physician dyads. The increased prevalence of other risk factors for BMA discharge among male patient-female physician dyads (e.g., male sex, prior BMA discharges, substance misuse) might explain this observation.

A third explanation suggests male patients unconsciously exhibit discriminatory attitudes toward female physicians, negatively impacting the patient-provider relationship and contributing to higher-than-expected rates of BMA discharge^32,33. Males have historically been vastly over-represented among physicians, and perhaps unconsciously female physicians are not automatically afforded the same levels of trust and reverence by both male and female patients. Male patients might also prefer a male physician because of in-group biases, embarrassment during the physical exam, or unsubstantiated assumptions that a male physician would be more empathetic to their particular concerns^34–36. Aggressive behavior among patients determined to leave BMA might be particularly threatening when the patient is male and the physician is female, reducing opportunities for de-escalation, negotiation, and counseling against an early discharge³⁷.

A fourth explanation posits our findings are the result of residual confounding. In our cohort, 80% of index hospitalizations had a male Most Responsible Physician, reflecting the over-representation of males in many hospital-based medical specialties (e.g., orthopedics, neurosurgery, cardiology; eFigures 1 and 2)^38–41 Moreover, female physicians may be better represented in medical specialties where BMA discharge is far more common (e.g., psychiatry, general internal medicine, addiction medicine), potentially explaining the increase in BMA discharge among male patient-female physician dyads (eFigures 1 and 2). Furthermore, because male patients are far more likely to choose BMA discharge and because only female physicians can be sex-discordant with male patients, residual confounding might generate a spurious association between sex discordance and BMA discharge that is unfairly attributed to female physicians. Future research with patient and physician interviews might help clarify the underlying mechanisms.

Policies designed to enforce patient-physician sex concordance are neither feasible nor ethical, and misguidedly encouraging patient-physician sex concordance may entrench existing sexist attitudes and communication failures¹⁰. Clinicians and decision-makers should instead focus on evidence-based interventions that could prevent BMA discharge and its harms. These interventions include adequate treatment of pain and opioid withdrawal^42,43; involvement of specialist addiction medicine consult teams for patients with substance use disorder^44,45; development of non-restrictive visitor and temporary pass policies⁴⁶; training clinicians in clinical negotiation and de-escalation; and the development of post-BMA outreach programs to prescribe needed medications, schedule follow-up tests and appointments, and re-engage community-based clinicians and social support organizations.

Our study has several strengths. Our study explores a previously unstudied association, finds empirical support for previously reported risk factors for BMA discharge, highlights evidence-based interventions to prevent or reduce harm after BMA discharge, and advances research on patient-physician dynamics. We used a large population-based sample, making results highly generalizable. Outcome misclassification is unlikely because medicolegal considerations motivate hospitals to carefully document BMA discharges. We used multilevel models that accounted for differences in patient, physician, and hospital characteristics. We adjusted for many relevant confounders. We used objective data to examine clinically relevant outcomes. We assessed the robustness of our primary findings through additional analyses.

Some study limitations relate to exposure. Hospitalized patients typically receive medical care from a team of clinicians (including multiple attending physicians and trainees), making it difficult to attribute patient outcomes to a single physician. For some patients, the Most Responsible Physician may be different from the physician responsible for the patient on the day of discharge. In some cases, a clinical interaction with the discharge-day physician can substantially influence a patient’s decision to depart hospital, but in other cases, the departure occurs before the discharge-day physician has completed their rounds. Taking these nuances into account, we focused on the Most Responsible Physician, reasoning they had the greatest cumulative impact on patient outcomes because they likely had the greatest number of encounters with the patient and the most opportunity to establish an empathetic therapeutic relationship. Exposure misclassification arising from team-based care or from discrepancies between discharge-day and Most Responsible Physician would bias effect estimates to the null, suggesting our results may underestimate the true impact of patient-physician sex discordance.

Our study has other limitations. For both patients and physicians, our data was limited to binary classification of biological sex (“female,” “male”) and lacked information on sociocultural gender (e.g., “woman,” “man,” “transgender”) and sexual identity. We also lacked data on homelessness and race, which could influence patient-physician interactions and BMA discharge⁴⁷. We lacked data on which patients were unable to initiate a BMA discharge because of involuntary hospitalization under the BC Mental Health Act. The proxy for physician clinical experience (years since medical school graduation) does not account for experience with hospital-based care, clinical hours per week, or leaves of absence. We lacked granular clinical details and relied on administrative data to identify comorbidities such as substance use disorder. Future research might address these gaps and consider whether patient-physician sex discordance influences outcomes in other contexts (e.g., transitions of care, transfers to long-term care), in specific disorders, or in other cultural settings.

In a retrospective cohort study examining 1.9 million hospitalizations, we found that patient-physician sex discordance was associated with a small increase in the risk of BMA discharge among male patients, but not among female patients. Based on these findings, hospital-based clinicians should be mindful of the importance of patient-physician sex discordance and its potential to influence meaningful patient outcomes, researchers should continue to evaluate the potential mechanisms by which sex discordance might influence engagement with the healthcare system, and hospitals should focus their efforts on implementing interventions known to reduce BMA discharge and its harms.

Supplementary Information

Below is the link to the electronic supplementary material.

Author Contribution:

JAS and MK were responsible for study concept. All authors contributed to the study design. JAS and MK designed the analytic strategy. JAS was responsible for acquisition of the data. MK had full access to all study data and was responsible for the integrity of the data and the accuracy of the data analysis. YY and DD contributed to the data analysis. JAS and MK were responsible for drafting the manuscript. All authors were responsible for critical revisions of the manuscript.

Funding

This study was supported by the Canadian Institutes of Health Research (grant numbers PJT-180343 & PJT-183955), the Vancouver Coastal Health Research Institute Innovation and Translational Research Award (AWD-017961), and the UBC Division of General Internal Medicine. JAS was supported by a Health Professional-Investigator Award from Michael Smith Health Research BC. Funding organizations were not involved in the design and conduct of the study; collection, management, analysis, and interpretation of the data; or preparation and review of this abstract.

Data Availability

Access to data provided by the Data Stewards is subject to approval, but can be requested for research projects through the Data Stewards or their designated service providers. All inferences, opinions, and conclusions drawn are those of the authors and do not reflect the opinions or policies of the Data Stewards.

Declarations:

Conflict of Interest:

The authors declare that they do not have a conflict of interest.