Study site and participants
The study was conducted at a Veterans Affairs Medical Center hospital affiliated with a major university medical school (the "intervention site") in the Midwestern United States. The Veterans Affairs' Institutional Review Board and the Research & Development Committee and the Subcommittee on Human Studies approved the study. All general medicine physicians who attended at the intervention site during the period July 1, 1999 through June 30, 2001 (68 senior residents and 74 attending physicians) were included in the study.
Study period
The pre-profiling and profiling periods of the study were conducted in two phases over a two-year period. The pre-profiling period took place from July 1, 1999 through June 30, 2000. The profiling period included the period from July 1, 2000 through June 30, 2001. During both phases of the study, all attending physicians and senior residents on all four medical services of the intervention site were sent a short, self-administered questionnaire at the end of their ward month intended to measure their overall satisfaction during that month. During the profiling period, all attending physicians on the four medical services at the intervention site were informed that they would be profiled. At the beginning of each ward month, a brief meeting between the Chief of the Medical Service and the four attending physicians assigned to each medical team in which expectations regarding attending supervision, teaching, and documentation are discussed is held. Throughout the profiling period, this standard orientation included the following statement read by the Service Chief:
"As you may know, the VA, at a national level, is exploring various methods of reducing patient length of stay. One proposed method involves the profiling of attending physicians. All medical ward services will be profiled this month. By profiling, I mean that your patients' lengths of stay will be compared to the patient length of stay of other attendings. We will, of course, use patient risk-adjustment methods when comparing your team's performance to other teams' data. At the end of the month, you and your team will then be given a short questionnaire in which your attitudes about profiling will be assessed. All of your responses will be kept confidential."
Each attending physician was also sent a letter from the Service Chief reminding him or her that they were being profiled.
Survey instruments
We developed three self-administered questionnaires for this study: one for senior residents (8 questions), one for attending physicians administered during the pre-profiling phase (14 questions), and a second for physicians who attended during the profiling phase (23 questions). The questionnaires were largely comprised of questions conceived of by the study authors. The questions are specific to content domains identified by the study authors as important and appropriate to the study theme. We reproduce the exact wording of all key questions in the results that follow. The senior residents' questionnaire was designed to assess: 1) their satisfaction during their ward month; 2) their perception of the extent to which their attending physicians were involved in patient care decisions; and 3) the degree of autonomy in making patient care decisions allowed them by their attending physician. The same senior resident questionnaire was used during both the pre-profiling and profiling phases. Attending physician questionnaires administered during the pre-profiling phase were designed to assess: 1) their satisfaction during their ward month; 2) how they thought they would react to being profiled; 3) their feelings concerning economic aspects of patient care [24]; and 4) their perception of the quality of care they provided. Questionnaires mailed to attending physicians during the profiling phase differed from the pre-profiling attending physician questionnaire in that respondents were also asked to indicate if being profiled caused them to decrease the ordering of tests and/or procedures, decrease their patients' lengths of stay, whether they felt pressured to discharge a patient from the hospital prematurely, and whether they were more involved than usual in the care of their patients.
Medical information systems data
We used the Department of Veterans Affairs' Patient Treatment File for July 1, 1998 through June 30, 2002. The Patient Treatment File contains a standardized hospital discharge abstract describing each patient's demographic (e.g., age, sex, race) and clinical characteristics (ICD-9-CM diagnosis and procedure codes). Diagnoses include the primary diagnosis and up to nine secondary diagnoses. We retrieved Patient Treatment File data for the intervention site for the pre-profiling phase (July 1, 1999 through June 30, 2000) and the profiling phase (July 1, 2000 through June 30, 2001). For the purpose of comparison to the study period, we obtained data for the year preceding the pre-profiling phase (July 1, 1998 through June 30, 1999) and for the year following the profiling phase (July 1, 2001 through June 30, 2002). Also for comparison purposes, we analyzed data for six control hospitals in the same VA hospital network as the intervention site. The total study sample included 9,307, 9,250, 9,319, and 9,633 admissions for the years July 1, 1998 through June 30, 2002, respectively. At the intervention site, the pre pre-profiling year (July 1, 1998 through June 30, 1999), the pre-profiling phase (July 1, 1999 through June 30, 2000), the profiling phase (July 1, 2000 through June 30, 2001), and the post-profiling year (July 1, 2001 through June 30, 2002) included 2,186, 1,982, 2,145, and 2,162 admissions, respectively.
Data analysis
Survey data
We tested for statistically significant differences in physician responses between phases using the Student's t test for continuous variables and chi square analysis for categorical variables, with a criterion of p < 0.05. We next tested for statistically significant differences using logistic regression to control for possible confounding effects related to physician characteristics, such as years since graduation from medical school, type of work planning to pursue after completion of residency (primarily outpatient- or inpatient-based general internal medicine, or subspecialty fellowship), and sex of senior residents. For attending physicians, independent variables included years since graduation from medical school, years attending at the hospital, primary specialty, and subspecialty. For the 5-item Likert scale response sets that ranged from "strongly disagree" to "strongly agree," from "very positive" to "very negative," and from "a great deal" to "not at all," we trichotomized responses (e.g., "disagree/strongly disagree," "neither agree nor disagree," and "agree/strongly agree").
Medical information system data
The main dependent variable of interest was hospital length of stay. The principle independent variable of interest was study phase (pre-profiling versus profiling). We tested for a change in length of stay associated with profiling occurring above and beyond the temporal trends in length of stay. The distribution of hospital length of stay is heavily skewed with a long right tail and some very high length of stay outliers. Such extreme outliers usually represent medically unique patients and we thus truncated length-of-stay outliers, defined as those patients with length of stay greater than 2 standard deviations above the mean length of stay [8]. In addition, we tested for differences in length of stay between the pre-profiling and profiling phases using robust regression, which uses a pseudovalues method that does not assume a normal distribution of the dependent variable [25, 26]. Comorbidity was measured using a validated approach proposed by Elixhauser and colleagues [27]. This approach uses secondary ICD-9 codes and considers 30 diagnoses. Algorithms exclude conditions that are related to the admission diagnosis and are unlikely to represent complications of care [27]. We also adjusted for patient demographics (age, marital status, race, and sex), discharge destination, and primary diagnosis.