Gender Differences in Compensation in Anesthesiology in the United States: Results of a National Survey of Anesthesiologists : Anesthesia & Analgesia

Secondary Logo

Journal Logo

Original Research Articles: Original Clinical Research Report

Gender Differences in Compensation in Anesthesiology in the United States: Results of a National Survey of Anesthesiologists

Hertzberg, Linda B. MD*; Miller, Thomas R. PhD, MBA; Byerly, Stephanie MD; Rebello, Elizabeth MD§; Flood, Pamela MD*; Malinzak, Elizabeth B. MD; Doyle, Christine A. MD; Pease, Sonya MD, MBA#; Rock-Klotz, Jennifer A. MBA; Kraus, Molly B. MD**; Pai, Sher-Lu MD††

Author Information
Anesthesia & Analgesia 133(4):p 1009-1018, October 2021. | DOI: 10.1213/ANE.0000000000005676

Abstract

KEY POINTS

  • Question: Does a gender compensation difference still exist among anesthesiologists?
  • Findings: Compensation for anesthesiologists continues to show a significant pay gap associated with gender even after adjusting for differences in age, hours worked, geographic practice region, and other variables.
  • Meaning: Bias, either explicit or implicit, persists and affects compensation for anesthesiologists.

A compensation gap associated with gender has been well documented among physicians and other professionals in the United States. The difference in salaries between men and women physicians has been persistent, even after accounting for factors such as age, experience, specialty, work hours, productivity, and academic rank.1–5 A recent study of US public medical schools reported that women physicians earned $19,879 or 8% less than men.5 In addition, compensation gaps associated with gender have been recognized across various specialties.6–11 Such gaps typically widen and persist over the entire course of a physician’s career.12

Few studies have assessed the compensation gap specifically among anesthesiologists. A 2007 study showed that Caucasian women anesthesiologists made $60,337 or 20% less than Caucasian men anesthesiologists, after adjusting for annual hours worked, provider characteristics, and practice characteristics.13 A 2013 survey, titled “Regional and Gender Differences and Trends in the Anesthesiologist Workforce,” was commissioned by the American Society of Anesthesiologists (ASA) through the RAND Corporation.14 The unadjusted gender wage gap in that survey was 29%. After adjusting for hourly earnings, it was 15%. After further adjustments for work experience and practice characteristics, the hourly earning gap was 7%.14 A 2017 study that evaluated Medicare Provider Utilization and Payment Data reported that women anesthesiologists had lower average total Medicare payments compared with men anesthesiologists.15

In 2018, the ASA’s Ad Hoc Committee on Women in Anesthesia (AHCWIA) was tasked to examine if a gender compensation gap still existed and to identify actionable items that might lessen the gap in the future. In response, the AHCWIA and the Center for Anesthesia Workforce Studies developed a survey, titled “Compensation Patterns in Anesthesiology,” to determine differences in compensation between men and women anesthesiologists and what potential factors contributed to that gap. The primary explanatory variable of interest was gender. Secondary variables that could potentially explain a compensation gap were identified by a literature search and a workgroup of the AHCWIA.

METHODS

Survey Development and Distribution

The study was deemed exempt by the Mayo Clinic institutional review board. A scoping literature review of US gender compensation gaps in medicine was conducted to guide the design of the survey questionnaire.16 PubMed was searched using the terms “compensation,” “difference,” “disparity,” “earnings,” “gap,” “gender,” “income,” “medicine,” “men,” “negotiation,” “physician,” “provider,” “reimbursement,” “salary,” “sex,” “women,” and “workforce” for English-language publications from 1996 to 2019 (Supplemental Digital Content 1, Box 1, https://links.lww.com/AA/D616). Reference lists of these published reports were reviewed to identify additional relevant reports and supplementary content. An analogous search in Google Scholar was conducted to identify gray literature, editorials, and opinion pieces. Two researchers (T.R.M., J.A.R.-K.) screened >2000 titles and abstracts to assess eligibility for inclusion; of these, 83 articles were deemed relevant to the research topic (52 peer-reviewed articles, 17 gray literature documents, and 14 editorials or opinion statements).

Although the 2013 RAND survey was not developed specifically to examine the relationship of compensation and gender, our scoping review and examination of the original survey tool indicated that it was a useful template for designing a survey to identify compensation patterns in anesthesiology.14 Our survey included questions on demographic characteristics, compensation, benefits, and factors affecting job search (Tables 1 and 2). These factors were based on the literature available in 2018 regarding gender effects on compensation and the expertise and opinions of the AHCWIA workgroup members.

Table 1. - Compensation and Demographic Characteristics of the Study Sample by Gender
Characteristic Womena (n = 686) Mena (n = 1395) P b Standardized differencec
Independent variable (ordinal value and compensation range)
Reported compensation range, % (CI)
 1. <$250,000 14 (11–16) 5 (3–6) <.001 0.729
 2. $250,000−$299,999 14 (11–16) 6 (4–7)
 3. $300,000−$349,999 24 (21–27) 14 (12–16)
 4. $350,000−$399,999 22 (19–25) 21 (19–24)
 5. $400,000−$449,999 16 (13–19) 22 (19–24)
 6. $450,000−499,999 6 (4–8) 13 (11–15)
 7. >$500,000 5 (3–7) 20 (18–22)
Demographic characteristics
 Race other than Caucasian, % (CI) 26 (23–29) 20 (18–22) .003 0.135
 Age, y, median (IQR) 41 (37–53) 45 (38–57) <.001 −0.300
Marital and work status, % (CI)
 Married—spouse works 60 (56–64) 42 (39–44) <.001 0.569
 Married—spouse does not work 24 (21–28) 50 (47–52)
 Single 16 (13–18) 9 (7–10)
Dependent children in household, % (CI) 52 (48–55) 60 (58–63) <.001 −0.176
Dependent adults in household, % (CI) 15 (12–18) 13 (11–14) .170 0.063
Region of United States, % (CI)
 South Atlanticd 21 (18–24) 20 (18–22) .045 0.259
 East North Central 13 (11–16) 17 (15–19)
 East South Central 5 (4–7) 6 (4–7)
 Mid-Atlantic 14 (11–16) 13 (11–14)
 Mountain 6 (4–8) 9 (7–10)
 New England 6 (5–8) 6 (4–7)
 Pacific 17 (15–20) 13 (12–15)
 West North Central 7 (5–8) 7 (6–9)
 West South Central 11 (8–13) 10 (9–12)
Trained locally, % (CI) 48 (45–52) 39 (36–41) <.001 0.190
Abbreviations: CI, confidence interval; IQR, interquartile range.
aContinuous data (age in years) are reported as median and IQR (first quartile to third quartile). Categorical data are reported as a percentage and CI (95% CI around mean).
bP values indicate whether the difference between the groups was statistically significant. Differences in age, years in practice, and weekly hours were assessed by the 2-sample Wilcoxon rank-sum test and a nonparametric equality of medians test. The χ2 test was used for the remaining variables.
cStandardized difference = difference in means or proportions divided by the pooled standard deviation.17 Absolute values <0.2 represent small differences; values 0.2 to 0.5 represent moderate differences; values >0.5 represent large differences.
dIncludes 10 men and 2 women respondents indicating practice location in a US territory.

Table 2. - Salary-Related Characteristics and Job Search Skills of the Study Sample by Gender
Characteristic Womena (n = 686) Mena (n = 1395) P b Standardized differencec
Salary and benefits
 Years in practice, median (IQR) 12 (8–22) 15 (8–27) <.001 −0.223
 Hours worked (weekly), median (IQR) 56 (47–60) 57 (50–64) .001 −0.249
 Takes call, % (CI) 78 (74–81) 79 (77–82) .306 −0.047
Practice type, % (CI)
 Academic 37 (33–40) 22 (20–24) <.001 0.461
 Hospital used 14 (11–17) 17 (15–19)
 Small group 9 (7–11) 5 (4–6)
 Large group 12 (9–14) 13 (12–15)
 Partner in practice 14 (12–17) 29 (27–32)
 Other 15 (12–17) 14 (12–15)
Academic rank, % (CI)
 No academic rank 62 (59–66) 72 (70–75) <.001 0.236
 Assistant professor 25 (22–28) 16 (14–18)
 Associate professor 8 (6–11) 7 (5–8)
 Full professor 4 (3–6) 5 (3–6)
Completed fellowship, % (CI) 48 (44–51) 40 (37–42) .001 0.154
Important factors in job selectiond, % (CI)
 Salary 27 (24–30) 44 (41–47) <.001 −0.357
 Less call or less weekends 32 (28–35) 26 (24–28) .007 0.124
 Flexible scheduling 28 (24–31) 17 (15–19) <.001 0.267
 Dependent caree 16 (13–19) 6 (5–8) <.001 0.316
 Leadership opportunities 13 (10–15) 14 (13–16) .244 −0.055
Paid family leave (taken), % (CI) 6 (4–7) 5 (4–6) .475 0.033
Paid time off (<25 d), % (CI) 40 (36–43) 32 (30–34) .001 0.156
Unpaid time off (none), % (CI) 36 (32–39) 42 (39–44) .012 −0.118
Job search skills
 Consulted a mentor, % (CI) 52 (48–55) 49 (46–52) .270 0.051
 Knew a person in the practice, % (CI) 64 (60–67) 59 (57–62) .045 0.094
 Negotiated salary or contract, % (CI) 19 (16–22) 25 (23–27) .002 −0.147
 Lawyer reviewed the contract, % (CI) 27 (23–30) 33 (31–36) .002 −0.146
Abbreviations: CI, confidence interval; IQR, interquartile range.
aContinuous data (years in practice and hours worked) are reported as median and IQR (first quartile to third quartile). Categorical data are reported as a percentage and CI (95% CI around mean).
bP values indicate whether the difference between the groups was statistically significant. Differences in age, years in practice, and weekly hours were assessed by the 2-sample Wilcoxon rank-sum test and a nonparametric equality of medians test. The χ2 test was used for the remaining variables.
cStandardized difference = difference in means or proportions divided by the pooled standard deviation.17 Absolute values <0.2 represent small differences; values 0.2 to 0.5 represent moderate differences; values >0.5 represent large differences.
dSurvey respondents were asked to select the top factors that were most important in choosing their job.
eChild- or other-dependent family care considerations.

Annual compensation-related data were reported either as a point estimate or in incremental ranges of $50,000. Initial sign-on bonuses, loan payoffs, and stipends for call, medical direction, or other administrative work were included. A range of employment and physician practice information was collected, including the type of practice (academic, private, government, locum tenens, and national group practice company), number of years in practice, mode of practice in the primary facility, and financial structure of the primary facility.

The survey was revised 3 times during its development and testing with the AHCWIA workgroup for consistency, accuracy, and ease of completion. A pilot survey was then distributed to members of the AHCWIA, the Committee on Young Physicians, and the Committee on Professional Diversity within the ASA. The pilot survey was returned by 61 of 120 physicians (50.8% response rate). The final survey included modifications made after reviewing the feedback and analyzing the pilot results for completion and consistency.

The “Compensation Patterns in Anesthesiology” survey was electronically distributed via e-mail using SurveyMonkey (SVMK Inc) to 28,812 ASA members on September 2, 2018. The survey was distributed to all active ASA members who had not opted out of receiving research surveys. Internet protocol addresses were not collected via the survey to ensure anonymity of the respondents. Active members were US anesthesiologists, excluding residents and fellows. Every other week, reminders were sent to participants until the close of the survey on October 29, 2018. To increase the survey response rate, members who successfully completed the survey could participate in a random drawing to win 1 of 10 e-gift cards, each valued at $500. Winners were confirmed through e-mail correspondence before they received the e-gift cards. The submitted e-mail addresses were separated from the data for analysis.

Statistical Analysis

Outcome Measure

Compensation was the primary outcome, defined as the amount reported as direct compensation on a W-2, 1099, or K-1, plus all voluntary salary reductions (eg, 401[k], health insurance). The survey was designed to direct the respondents to include salary, bonuses, incentive payments, research stipends, honoraria, and distribution of profits to employees. However, for this study, profits resulting from corporate ownership and the value of benefits paid by the practice (eg, retirement plan contributions, malpractice premiums, and health insurance) were not categorized as compensation.

To increase the survey response rate, respondents had the option of providing a point estimate of their compensation or selecting a range (in $50,000 increments). Approximately half of the survey respondents provided a point estimate of their compensation, and the remainder chose the range options.

Model Variables

Gender was the explanatory variable of interest. The following items were included as potentially confounding variables in the regression models: race, marital and work status, dependent children or adults, US Census region, trained locally, years in practice, weekly hours worked, taking call, practice type, academic rank, completed fellowship, factors considered important in job selection (ie, salary, less call or less weekends, flexible scheduling, dependent care, and leadership opportunities), paid family leave, paid time off <25 days, no unpaid time off, consulted a mentor, knew a person in the practice, negotiated salary or contract, and lawyer review of contract.

Restricted cubic splines were applied to the continuous variables age, weekly hours, and years in practice.18,19 All other variables were coded as binary indicators. Tables 1 and 2 compare the outcome and model variables between men and women. Continuous data were reported as median (first quartile to third quartile) and binary indicators as a number (%). Since P values are influenced by sample size, the standardized difference between men and women for each variable was reported to help identify important differences. The standardized difference is the difference in means or proportions divided by the pooled standard deviation.17 Absolute values <0.2 represented small differences, values 0.2 to 0.5 represented moderate differences, and values >0.5 represented large differences. In the Results section, we identified variables for which differences between men and women in the analytic sample were statistically significant (P <.05) and represented moderate or large standardized differences.

SAS Enterprise Guide V7.15 (SAS Inc) was used to develop the analytic sample from the survey data and to calculate standardized differences. Stata/MP 16.0 (StataCorp LLC) was used for all other statistical analysis. The adequacy of the analytic sample size was assessed based on the number of outcome events relative to the number of predictor variables and the criteria proposed by Riley et al.20,21 In the model with the most predictor variables (49), there were 42 outcome events per predictor variable (2081/49).

Statistical Model

Multivariable ordinal logistic regression was used to estimate odds ratios and 95% confidence intervals for the relation between gender and compensation, with adjustment for potentially confounding covariates.22,23 Ordinal logistic regression (proportional odds regression) is an extension of logistic regression that is useful when the number of categorical outcomes is >2 and the categories follow in an ordered sequence. In the primary analysis, the ordinal variable was defined as the 7 ranges of compensation listed in Table 1. The 4 survey-based compensation ranges <$250,000 were combined because of the small sample sizes in the lower ranges.

The Wolfe-Gould likelihood ratio test and Brant Wald test were used to test the proportional odds assumption.24,25 Both tests indicated that our data breached the proportional odds (parallel regression) assumption (P < .001). A violation of this assumption indicated that the effects of one or more independent variables significantly vary across the compensation ranges, and the interpretation of the estimated odds ratios may not be valid. Seventeen potential confounders had effects that varied significantly across the compensation ranges. Importantly, the effect of the variable of interest and gender did not vary across the compensation ranges.

A generalized ordinal logistic regression, also known as a partial proportional odds model, was fitted to address the violation of the proportional odds assumption.22,26,27 For our generalized ordinal logistic regression model, the global Wald test (P = .228) indicated that our model did not violate the proportional odds assumption, demonstrating the model was good and sufficient.

The generalized model relaxes the requirement that all covariates meet the strict proportional odds (parallel lines) assumption required under the ordinal logistic regression model. Using the generalized model, the interpretation of the odds ratios was the same as in the strict ordinal logistic regression for those variables where the proportional odds assumption held, as with gender, the variable of interest in our model. The interpretation was simply the odds of being in a higher compensation range.

The interpretation of the results on variables where the proportional odds assumption is relaxed, however, is less straightforward. For each of the ordinal values (ie, compensation ranges in our model), there were different odds ratios for these variables, and they should be examined for each of the compensation range-specific logistic regressions. That is, for a given compensation range, the odds ratios associated with these variables represent the estimated odds of the compensation being in the specified compensation range versus being outside that range, whether higher or lower. Since the covariates other than gender were considered potential confounders, interpretation of their odds ratios was not necessary.

Sensitivity Analyses

Four linear regression models on the log-transformed compensation estimates were run as sensitivity analyses. For the subset of the respondents that only provided a compensation range, either the median or mean within each range from the sample subset with point estimates or the midpoint of the range was used as point estimates. The models were (1) a linear regression of the log-transformed compensation for only the sample subset that provided point estimates of compensation; (2) a similar model for the full sample, replacing the range given with the median of the range based on the sample subset with point estimates; (3) a model similar to model 2 but using the mean value within each range; and (4) a model similar to model 2 but using the midpoint of the range (eg, for respondents indicating a range of $300,000–$350,000, $325,000 was used). Additional models were run to test interaction terms, including practice type, years in practice, and the variable indicating salary was a very important factor in job selection. As a final sensitivity analysis, an ordinal logistic regression that ignored the violation of the proportional odds assumption was run to examine the association with the estimated odds ratio for gender equal to woman.

RESULTS

Survey Response and Analytic Sample

We received 3025 responses to the survey (response rate, 10.5%). From these, we excluded responses from residents, fellows, anesthesiologists not practicing in the United States, and duplicate and incomplete responses. We also excluded surveys that did not answer the questions about gender, compensation, and age. The final analytic sample consisted of 2081 observations (response rate, 7.2%). Women physicians provided 33.0% of responses. This percentage was higher than the percentage of women (25.7%) in the overall demographic characteristics of practicing anesthesiologists in the United States of the same year.28,29

Tables 1 and 2 present characteristics of the study sample by gender. Our sample population was younger than the median age of anesthesiologists from the American Medical Association Physician Masterfile (44 vs 53 years).30 In comparing men and women in the analytic sample, the independent variable (compensation range) and 20 of the potential confounders had statistically significant (P < .05) differences. However, only 9 potential confounders also had moderate standardized differences (absolute value between 0.2 and 0.5, inclusive), and only the independent variable and 1 potential confounder had a large standardized difference (>0.05). The differences between men and women for all 3 continuous variables—age, years in practice, and hours worked—were statistically significant and had moderate standardized differences. The binary variables with statistically significant differences and moderate or large standardized differences between men and women were: marital and work status, region of United States, practice type, academic rank, and the following important factors in job selection: salary, flexible schedule, and family or childcare considerations.

Results of the Generalized Ordinal Logistic Regression

The odds ratio associated with woman, the variable of interest, after adjusting for potential confounders, was an estimated 0.44 (95% confidence interval, 0.37–0.53). This indicates that for a given compensation range, women had an estimated 56% lower odds than men of being in a higher compensation range versus being at or below that given range. For example, for the compensation range $350,000 to $399,000, women had an estimated 56% lower odds than men of compensation of $400,000 or above versus <$400,000. Since the proportional odds assumption holds for the variable woman, this 56% lower odds applies to each of the first 6 compensation ranges. In ordinal logistic regression, the odds ratio does not apply to the highest ordinal value (ie, highest compensation range in our model, ≥$500,000).

F1
Figure.:
Predicted probabilities and 95% CIs for 7 ranges of reported compensation, by gender. Based on the generalized ordinal logistic regression, holding all other covariates at their means. CI indicates confidence interval.

The predicted probabilities, by gender, of being in each of the compensation ranges are presented in the Figure and Supplemental Digital Content 2, Figure 1, https://links.lww.com/AA/D617. Because interaction terms cannot be included in the generalized ordinal logistic regression, we plotted predicted probabilities, by gender, for selected characteristics (ie, practice type, years in practice, and salary identified as very important in job selection). These graphs are presented in Supplemental Digital Content 2, Figures 2a–c and 3a, b, https://links.lww.com/AA/D617. Based on visual inspection, we see a similar pattern of probabilities by gender across the characteristics displayed.

Results of the Sensitivity Analyses

Table 3 presents summary results of the 4 models based on linear regressions of log-transformed compensation estimates that include the same covariates as in the generalized ordinal logistic regression. The parameter estimates associated with women were within a narrow range; the relative percentage difference in compensation for women compared to men ranged from −8.3 to −8.9. For the subset of respondents who provided a point estimate of compensation (model 1, n = 1036), the −8.3% reflects a reduction in compensation for women, holding other covariates at their means, of $32,617. Over a 30-year career, this could represent almost a million-dollar shortfall. For the sensitivity analyses with linear regression models that included interaction terms (not shown), none of the interactions was statistically significant.

Table 3. - Sensitivity Analyses: Results of Linear Regressions of Log-Transformed Compensation
Model number and description Percentage difference in compensation for women compared to mena 95% confidence intervals
 1. Based on sample of respondents providing point estimates of compensation (n = 1036) −8.3 −4.7 to −11.7
Models with imputed values of compensation for respondents providing only a rangeb, based on:
 2. Mean values within relevant range −8.9 −6.5 to −11.1
 3. Median values within relevant range −8.5 −6.3 to −10.7
 4. Based on range midpoint for ordinal intervals 2 to 6c and median values for ordinal intervals 1 and 7d −8.5 −6.2 to −10.6
aThe associated P value for this parameter estimate in each model is <.001. The adjusted R2 for each model is 0.36.
bMeans and medians were calculated within each compensation range (ordinal interval) for the 1036 respondents that provided point estimates of compensation. There are 1045 imputed observations in these models, for a total of 2081 observations.
cFor example, the range midpoint for ordinal interval 2 = $275,000 (halfway between 250,000 and 300,000).
dOrdinal interval 1 ≤ $250,000; ordinal interval 7 = ≥$500,000. As in model 3, the median values for the 1036 respondents that provided point estimates of compensation were used for these 2 intervals.

The ordinal logistic regression sensitivity analysis that ignored the violation of the proportional odds assumption had similar results as our generalized ordinal logistic regression. The adjusted odds ratio associated with woman, the exposure variable of interest, was an estimated 0.46 (95% confidence interval, 0.38–0.55).

DISCUSSION

We report lower compensation for women than for men after adjusting for potential confounding variables. These survey results are consistent with published articles reporting gender differences in compensation among physicians. Based on predicted probabilities within compensation ranges, the gender differences appeared consistent across several subgroups in our sample: practice type and years in practice (Supplemental Digital Content 2, Figures 2a–c and 3a, b, https://links.lww.com/AA/D617).

In 2018, 35.8% of all physicians in the United States were women, and the percentage of women anesthesiologists was 25.7%.28,29,31 If lack of gender parity in anesthesiology is the explanation, we would expect medical disciplines with greater gender parity to have pay equity. Yet, substantial compensation gaps exist in obstetrics-gynecology, geriatric medicine, pediatrics, and other specialties with a higher proportion of women physicians.8,10,15 A recent study of US physician income (2014–2018) showed that although practices and specialties with higher proportions of women physicians had a lower income gap, a gap still existed and was exaggerated in both the surgical specialties and in practices with >90% men.32 Another study noted that as the proportion of women rose in a medical specialty, overall compensation in that specialty fell, with persistence of the gender pay gap.33 Although the average compensation of anesthesiologists was found to be higher than the average of physicians in all specialties, the gender-based compensation gap remained.

The median age and years in practice were greater for men than women in anesthesiology. This difference can be explained by the gender distribution by age among anesthesiologists, with a lower percentage of women in the older age groups. It may also suggest a phenomenon of attrition for midcareer women physicians. This study was unable to capture anesthesiologists who had left the field of medicine. The surveyed population likely comprised anesthesiologists who remained in medicine due to contextual factors such as supportive spouse, equity in leadership opportunities, and/or freedom from workplace harassment and discrimination.

In our study, men anesthesiologists reported their weekly median work hours to be 4 hours higher than women anesthesiologists. They were also twice as likely to be the sole wage earners with spouses who did not work outside the home. Men may benefit from the questionable idea that men physicians supporting a family should receive a higher salary than women physicians who are not the sole financial providers for their households.3 However, these considerations do not seem to account for the substantial gender compensation gap that remained after adjusting for hours worked.

We found that women who were partners in practice settings had lower compensation (Supplemental Digital Content 2, Figure 2b, https://links.lww.com/AA/D617) than men after adjusting for most of the variables considered to affect compensation. We hypothesize that leadership and partnership positions may not be automatically given when a physician reaches milestones such as number of years in practice, cases performed, revenue generated, or administrative activities performed. A partnership offer may be based on subjective rather than objective factors. Skills such as self-promotion and negotiation, which are perceived as strengths for men, are historically perceived negatively for women.34,35 This suggests that an inequality in compensation exists, even for physicians who are considered “equal” partners.

A considerable percentage of academic women anesthesiologists who responded to our survey held the rank of assistant professor. This finding was consistent with prior studies that have shown that women medical school faculty neither advances as rapidly nor is paid at the same level as professionally equivalent men colleagues.1,36–38 Academic women physicians have midcareer research productivity rates that are equivalent to or greater than those of men physicians, with women often showing increased productivity after childbearing age and surpassing publication rates of men.39 However, because men often have greater research productivity earlier in their careers, they are more likely to be placed earlier in the “advancement track” by academic promotion committees.39 For women assistant professors who reached similar productivity to men by midcareer, publication volume often has no meaningful effect on their rank.40 We attribute the attrition of women physicians due to gender bias in initial compensation and in subsequent academic promotions to be the most likely explanation underlying the gender compensation gap in academic medicine.

Our study has several limitations. Anesthesiologists who were not ASA members or who have left the work force were not surveyed. Gender was self-identified as a binary variable. The study presents self-reported information and is a single, cross-sectional, survey-based study that potentially has response bias. Compared with the overall demographic characteristics of anesthesiologists in the United States, a larger proportion of our survey respondents were women (25.7% vs 33.0%).28,29 The analytic sample was not representative of the years of experience and age of anesthesiologists in 2018 overall. In that year, the mean age of anesthesiologists was 51.5 years,30 which was 4.7 years older than the mean age of respondents in the analytic sample. The survey did not identify the employer of the respondent. Therefore, individual group or institutional differences could not be estimated and adjustments for possible correlation among physicians used by the same group or hospital could not be made. Additionally, it was beyond the scope of this study to examine productivity in a manner other than the measured hours of work. However, anesthesia groups that use anesthesia units or modified units to measure productivity and provide compensation still have risk of gender bias. A compensation disparity could be attributable to case assignments, physician subspecialties, surgeon preferences, and case compensation mix, even when the hours worked are equivalent.

Our findings reinforce that a gender compensation gap exists even after adjusting for the physician and practice characteristics that influence compensation. Bias, either explicit or implicit, persists and affects compensation for women anesthesiologists. A “Statement on Compensation Equity Among Anesthesiologists” (Supplemental Digital Content 3, Box 2, https://links.lww.com/AA/D618) was adopted by the ASA House of Delegates in October 2019. Practices and other organizations using anesthesiologists should implement the recommendations in that document. We plan to repeat the survey in 5 years to examine the effects of the recommendations and to identify changes that may have occurred in anesthesiologists’ compensation related to gender. Any future study should also include race and ethnicity as primary variables of interest.

ACKNOWLEDGMENTS

We thank James D. Grant, MD, MBA, 2018 President of the American Society of Anesthesiologists (ASA), Senior Vice President and Chief Medical Officer, Blue Cross Blue Shield of Michigan, Bloomfield Hills, MI, for the staffing and financial support that he authorized via ASA to the Ad-Hoc Committee on Women in Anesthesia for this project. We also thank Eric C. Sun, MD, PhD, Assistant Professor of Anesthesiology, Perioperative and Pain Medicine, Stanford University Medical Center, Stanford, CA, and Mike Hernandez, MS, Senior Biostatistician, Biostatistics Department, The University of Texas MD Anderson Cancer Center, Houston, TX, for their assistance with the statistical analysis.

DISCLOSURES

Name: Linda B. Hertzberg, MD.

Contribution: This author helped with study conception and design, acquisition of data, analysis and interpretation of data, drafting and revising the manuscript, and final approval of the manuscript.

Name: Thomas R. Miller, PhD, MBA.

Contribution: This author helped with study conception and design, acquisition of data, analysis and interpretation of data, drafting and revising the manuscript, and final approval of the manuscript.

Name: Stephanie Byerly, MD.

Contribution: This author helped with interpretation of data, drafting and revising the manuscript, and final approval of the manuscript.

Name: Elizabeth Rebello, MD.

Contribution: This author helped with interpretation of data, drafting and revising the manuscript, and final approval of the manuscript.

Name: Pamela Flood, MD.

Contribution: This author helped with interpretation of data, drafting and revising the manuscript, and final approval of the manuscript.

Name: Elizabeth B. Malinzak, MD.

Contribution: This author helped with study conception and design, and final approval of the manuscript.

Name: Christine A. Doyle, MD.

Contribution: This author helped with study conception and design, drafting and revising the manuscript, and final approval of the manuscript.

Name: Sonya Pease, MD, MBA.

Contribution: This author helped with study conception and design.

Name: Jennifer A. Rock-Klotz, MBA.

Contribution: This author helped with study conception and design, analysis and interpretation of data, drafting and revising the manuscript, and final approval of the manuscript.

Name: Molly B. Kraus, MD.

Contribution: This author helped with study conception and design, analysis and interpretation of data, drafting and revising the manuscript, and final approval of the manuscript.

Name: Sher-Lu Pai, MD.

Contribution: This author helped with interpretation of data, drafting and revising the manuscript, and final approval of the manuscript.

This manuscript was handled by: Zeev N. Kain, MD, MBA.

REFERENCES

1. Ash AS, Carr PL, Goldstein R, Friedman RH. Compensation and advancement of women in academic medicine: is there equity? Ann Intern Med. 2004;141:205–212.
2. DesRoches CM, Zinner DE, Rao SR, Iezzoni LI, Campbell EG. Activities, productivity, and compensation of men and women in the life sciences. Acad Med. 2010;85:631–639.
3. Jagsi R, Griffith KA, Stewart A, Sambuco D, DeCastro R, Ubel PA. Gender differences in the salaries of physician researchers. JAMA. 2012;307:2410–2417.
4. Jagsi R, Griffith KA, Stewart A, Sambuco D, DeCastro R, Ubel PA. Gender differences in salary in a recent cohort of early-career physician-researchers. Acad Med. 2013;88:1689–1699.
5. Jena AB, Olenski AR, Blumenthal DM. Sex differences in physician salary in US public medical schools. JAMA Intern Med. 2016;176:1294–1304.
6. Ly DP, Seabury SA, Jena AB. Differences in incomes of physicians in the United States by race and sex: observational study. BMJ. 2016;353:i2923.
7. Weeks WB, Wallace AE. Association of race and gender with general surgeons’ annual incomes. J Am Coll Surg. 2006;203:558–567.
8. Weeks WB, Wallace AE. The influence of physician race and gender on obstetrician-gynecologists’ annual incomes. Obstet Gynecol. 2006;108:603–611.
9. Weeks WB, Wallace AE. Race and gender differences in pediatricians’ annual incomes. Ambul Pediatr. 2007;7:196–200.
10. Weeks WB, Wallace TA, Wallace AE. How do race and sex affect the earnings of primary care physicians? Health Aff (Millwood). 2009;28:557–566.
11. Weeks WB, Wallace AE. Race and gender differences in general internists’ annual incomes. J Gen Intern Med. 2006;21:1167–1171.
12. Bowles HR, Babcock L, McGinn KL. Constraints and triggers: situational mechanics of gender in negotiation. J Pers Soc Psychol. 2005;89:951–965.
13. Weeks WB, Wallace AE, Mackenzie TA. Gender differences in anesthesiologists’ annual incomes. Anesthesiology. 2007;106:806–811.
14. Baird M, Daugherty L, Kumar KB, Arifkhanova A. Regional and gender differences and trends in the anesthesiologist workforce. Anesthesiology. 2015;123:997–1012.
15. Mahr MA, Hayes SN, Shanafelt TD, Sloan JA, Erie JC. Gender differences in physician service provision using Medicare claims data. Mayo Clin Proc. 2017;92:870–880.
16. Munn Z, Peters MDJ, Stern C, Tufanaru C, McArthur A, Aromataris E. Systematic review or scoping review? Guidance for authors when choosing between a systematic or scoping review approach. BMC Med Res Methodol. 2018;18:143.
17. Yang D, Dalton J. A unified approach to measuring the effect size between two groups using SAS. SAS global forum. 2012. Accessed November 9, 2020. http://support.sas.com/resources/papers/proceedings12/335-2012.pdf.
18. Harrell FE Jr, Lee KL, Pollock BG. Regression models in clinical studies: determining relationships between predictors and response. J Natl Cancer Inst. 1988;80:1198–1202.
19. Riester KA, Peduzzi P, Holford TR, Ellison RT 3rd, Donta ST. Statistical evaluation of the role of Helicobacter pylori in stress gastritis: applications of splines and bootstrapping to the logistic model. J Clin Epidemiol. 1997;50:1273–1279.
20. Riley RD, Snell KI, Ensor J, et al. Minimum sample size for developing a multivariable prediction model: PART II - binary and time-to-event outcomes. Stat Med. 2019;38:1276–1296.
21. Riley RD, Snell KIE, Ensor J, et al. Minimum sample size for developing a multivariable prediction model: part I: continuous outcomes. Stat Med. 2019;38:1262–1275.
22. Williams R. Generalized ordered logit/partial proportional odds models for ordinal dependent variables. Stata J. 2006;6:58–82.
23. Williams R. Understanding and interpreting generalized ordered logit models. J Math Sociol. 2016;40:7–20.
24. Wolfe R, Gould W. An approximate likelihood-ratio test for ordinal response models. Stata Tech Bull. 1998;42:24–27.
25. Brant R. Assessing proportionality in the proportional odds model for ordinal logistic regression. Biometrics. 1990;46:1171–1178.
26. Fullerton AS, Xu J. The proportional odds with partial proportionality constraints model for ordinal response variables. Soc Sci Res. 2012;41:182–198.
27. Peterson B, Harrell FE. Partial proportional odds models for ordinal response variables. Appl Stat. 1990;39:205–217.
28. Association of American Medical Colleges. Diversity in Medicine: facts and figures 2019. Table 12: practice specialty, females by race/ethnicity. 2018. Accessed January 9, 2020. https://www.aamc.org/data-reports/workforce/data/table-12-practice-specialty-females-race/ethnicity-2018.
29. Association of American Medical Colleges. Diversity in Medicine: facts and figures 2019. Table 13: practice specialty, males by race/ethnicity. 2018. Accessed January 9, 2020. https://www.aamc.org/data-reports/workforce/data/table-13-practice-specialty-males-race/ethnicity-2018.
30. American Medical Association. AMA Physician Masterfile. Accessed January 2, 2019. https://www.ama-assn.org/practice-management/masterfile/ama-physician-masterfile.
31. Association of American Medical Colleges. Diversity in Medicine: facts and figures 2019. Figure 19: percentage of physicians by sex. 2018. Accessed January 9, 2020. https://www.aamc.org/data-reports/workforce/interactive-data/figure-19-percentage-physicians-sex-2018.
32. Whaley CM, Arnold DR, Gross N, Jena AB. Practice composition and sex differences in physician income: observational study. BMJ. 2020;370:m2588.
33. Pelley E, Carnes M. When a specialty becomes “women’s work”: trends in and implications of specialty gender segregation in medicine. Acad Med. 2020;95:1499–1506.
34. Rudman LA. Self-promotion as a risk factor for women: the costs and benefits of counterstereotypical impression management. J Pers Soc Psychol. 1998;74:629–645.
35. Stevens CK, Bavetta AG, Gist ME. Gender differences in the acquisition of salary negotiation skills: the role of goals, self-efficacy, and perceived control. J Appl Psychol. 1993;78:723–735.
36. Pololi LH, Civian JT, Brennan RT, Dottolo AL, Krupat E. Experiencing the culture of academic medicine: gender matters, a national study. J Gen Intern Med. 2013;28:201–207.
37. Fried LP, Francomano CA, MacDonald SM, et al. Career development for women in academic medicine: multiple interventions in a department of medicine. JAMA. 1996;276:898–905.
38. Richter KP, Clark L, Wick JA, et al. Women physicians and promotion in academic medicine. N Engl J Med. 2020;383:2148–2157.
39. Pashkova AA, Svider PF, Chang CY, Diaz L, Eloy JA, Eloy JD. Gender disparity among US anaesthesiologists: are women underrepresented in academic ranks and scholarly productivity? Acta Anaesthesiol Scand. 2013;57:1058–1064.
40. Chenoweth E, Fortna P, Mitchell S, Savun B, Weeks J, Cunningham K. How to get tenure (if you’re a woman). Foreign policy. Accessed December 10, 2019. https://foreignpolicy.com/2016/04/19/how-to-get-tenure-if-youre-a-woman-academia-stephen-walt/.

Supplemental Digital Content

Copyright © 2021 International Anesthesia Research Society