Characteristics of Women Internists

Erica Frank, MD, MPH; Tricia Kunovich-Frieze, MD; Giselle Corbie-Smith, MD

In This Article


The design and methods of the WPHS have been more fully described elsewhere, as have basic characteristics of the population.[3,4,5] WPHS surveyed a stratified random sample of US women MDs; the sampling frame is based on the American Medical Association's (AMA's) Physician Masterfile, a database intended to record all MDs residing in the United States and possessions. Using a sampling scheme stratified by decade of graduation from medical school, the authors randomly selected 2500 women from each of the last 4 decades' graduating classes (1950 through 1989). The authors oversampled older women physicians, a population that would otherwise have been sparsely represented by proportional allocation because of the recent increase in numbers of women physicians. The authors included active, part-time, professionally inactive, and retired physicians, aged 30-70 years, who were not in residency training programs in September 1993, when the sampling frame was constructed. In that month, the first of 4 mailings was sent out; each mailing contained a cover letter and a self-administered 4-page questionnaire. Enrollment was closed in October 1994 (final N = 4501).

Of the potential respondents, an estimated 23% were ineligible to participate because their addresses were wrong, they were men, they were deceased, they were living out of the country, or they were interns or residents. The response rate was 59% of physicians eligible to participate. The authors compared respondents and nonrespondents in 3 ways: (1) they used a phone survey (comparing a phone-surveyed random sample of 200 nonrespondents with all the written survey respondents), (2) they used the AMA Physician Masterfile (contrasting all respondents with all nonrespondents), and (3) they examined survey mailing waves (all respondents, from wave 1 through 4) to compare respondents and nonrespondents regarding a large number of key variables. From these 3 investigations, the authors found that nonrespondents were less likely than respondents to be board-certified. However, respondents and nonrespondents did not consistently or substantively differ on other tested measures, including age, ethnicity, marital status, number of children, alcohol consumption, fat intake, exercise, smoking status, hours worked per week, frequency of being a primary care practitioner, personal income, or percentage actively practicing medicine.

Based on these findings, the authors weighted the data by decade of graduation (to adjust for the stratified sampling scheme) and by decade-specific response rate and board-certification status (to adjust for the identified response bias). This allowed them to make inference to the entire population of women physicians graduating from medical school between 1950 and 1989.

From the 18 specialties reported, the authors created 4 categories of physicians. "General internists" were those declaring their specialty as general internal medicine with no subspecialty training. "Subspecialized internists" were those either listing medical subspecialist as their specialty or general internal medicine with subspecialty training. "Other specialists" included anesthesiologists, dermatologists, specialists in emergency medicine, neurologists, ophthalmologists, pathologists, psychiatrists, radiologists, all surgeons, and any marking "other." "Other primary care physicians" were defined as family practitioners, general practitioners, obstetrician/gynecologists, pediatricians, and public health physicians.

Contrasts of interest were: (1) general internists vs subspecialized internists, (2) general internists vs other primary care physicians, (3) subspecialized internists vs other specialists, and (4) all internists vs all others. The rules used for testing possible contrasts were as follows: first, the authors determined if there was any association (P ≥ .05) between specialty (4 categories) and the variable of interest. If so, the authors continued by testing contrasts 1, 2, and 3 above, using P ≤ .01 as the significance level of note. When there was heterogeneity either within the 2 internist groups or within the 2 noninternist groups, the authors did not test the last contrast (all internists vs all others).

To assess predictors of career satisfaction, the authors used logistic regression modeling. Satisfaction was dichotomized to: satisfied always/almost always/usually vs sometimes/rarely/never. Predictors initially offered were variables found to be associated with career satisfaction among all women physicians[6]; specifically, age, control of work environment, workplace stress, experience of harassment in a medical setting, number of days with poor mental health in the last month, work amount, and number of on-call nights per month. Model selection was by backward elimination, with a final inclusion criterion of P = .10 for the Wald F-test. The Hosmer-Lemeshow method[7] was used to test goodness of fit for each selected model. All tests were conducted using SUDAAN.[8]


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.
Post as: