We utilized the National Ambulatory Medical Care Survey (NAMCS),12
a nationally representative sample of non-hospital-based ambulatory care visits maintained by the National Center for Health Statistics, to identify adult patient visits to physicians in general internal medicine, family practice, general practice, and geriatrics from 1997 to 2005. NAMCS employs a multistage probability sampling design that samples all the ambulatory patient visits of office-based physicians engaged in direct care for 1 week randomly selected during the year. Physicians across specialties were sampled from 1 of 112 geographically based probability sampling units across the country. Actual visits sampled in each year were weighted accordingly. All of our analyses were adjusted for these weights in order to generate national estimates.
Physicians and their staffs were responsible for completing a visit record abstraction instrument for each visit that was separate from billing and reimbursement procedures. Each record included characteristics of the patient as well as physician and clinic characteristics. The survey documented visit diagnoses addressed using the International Classification of Diseases, Ninth Edition, Clinical Modification
medications prescribed using the National Drug Code Directory,14
diagnostic and screening tests ordered, various types of counseling provided, and physical therapy ordered. Diagnoses were limited to three in all years, the maximum documentable throughout the survey period. Medications were limited to six in all years. Later years allowed greater reporting of medications, but medications were limited to a maximum of six in all years to maintain consistency. All reports of exams, tests, counseling, and physical therapy were obtained from the use of checkboxes on the survey instrument. Only characteristics or activities that were consistently reported as checkboxes across all years were utilized, although in some cases similar categories were as collapsed into a single variable (Online Appendix Table
). In 2005, the survey instrument allowed documentation of 14 chronic diseases by using checkboxes, but this reporting was not utilized in this analysis. Other than a simple count of the diagnoses and medications, no write-in information was used.
The primary outcomes of interest were the number of clinical items addressed, visit duration (direct time spent with physician in minutes), and average available time per clinical item (duration/clinical items), measured in each year from 1997 to 2005. When calculating the total number of clinical items addressed, we assigned an equal weight for each diagnosis code (up to three, the maximum reported across all years), each medication (up to six, the maximum reported in early years), each diagnostic test (blood pressure, urinalysis, EKG, x-ray, mammography, other imaging, pregnancy test, pap smear, hematocrit or CBC, cholesterol, PSA, and other blood work), each act of counseling (diet, exercise, mental health or stress, and tobacco cessation), and physical therapy. Our measure of clinical items is an attempt to itemize the cognitive, logistical, and communication tasks that occur during the visit. For example, caring for a patient who presents with chest pain includes reviewing the differential diagnosis based on the history and physical exam, ordering diagnostic tests, and prescribing appropriate medications. Our measure of clinical items does not attempt to account for variation in the complexity of medical decision-making, challenges in patient-doctor communication, or administrative and clinical work performed outside of the visit. Visit duration was directly reported in NAMCS and represents the time that the physician directly spent with the patient. Average available time per clinical item for each visit was calculated by dividing visit duration by number of clinical items. The analysis was performed using nationally representative data on the average number of clinical items addressed, duration of visits, and available time per clinical item, derived from a sample of 46,431 adult primary care visits over the interval 1997–2005. To assess how these outcomes changed over time, we compared each subsequent year to the baseline year of 1997, allowing the pattern of change over time to be nonlinear.
In addition to assessing changes in all available visits, we also performed stratified analyses for visit by patients of different ages (18–39, 40–49, 50–64, 65–79, and ≥80) and by patients with different payer types (private insurance, Medicare, Medicaid, and self-pay), since age and insurance status were found to be significantly associated with complexity in all years. As well, in order to understand what clinical components contributed to changes in the overall volume of work associated with visits over time, we evaluated changes in the individual subcategories of clinical activity over time and evaluated their contribution to the total number of clinical items addressed per visit.
In addition, we performed adjusted analyses for each of the visit characteristics to account for potential secular changes in patient, physician, and clinic characteristics. Specifically, visit characteristics were adjusted for year of the patient visit, patient demographics (age, sex, race), payer type, urban setting for the clinic visit, geographic region of the clinic visit, whether the patient was seen by the patient’s regular primary care physician, whether the patient had been seen by the doctor before, whether the patient was referred to be seen by another physician, whether the physician is a solo practitioner, and ownership type of the clinic. Practice and payment characteristics were included in the model because they are plausibly related to visit duration, and similar variables have been shown to effect visit duration in past analyses.9
In all analyses, we used ordinary least squares regression models.
Lastly, to evaluate the relationship between duration and clinical items, we created an additional adjusted model of duration that also included the number of clinical items as a predictor. For this model, the number of clinical items was assumed to be exogenous to duration. More specifically, clinical items were assumed to derive from the medical issues associated with an individual patient and by the standard of care promulgated by the medical community to address those issues. To the extent that individual physicians to some degree induced more clinical items, we attempted to account for this possibility by adjusting for physician and clinic characteristics.
The primary outcome measures were not normally distributed and we conducted analyses with log-transformed outcomes. We found that log-transformation of outcome measures did not significantly alter our results in comparison to analyses with the original outcome measures. To enhance the interpretability of our analyses, we present results from analyses using the original untransformed outcome measures. All models utilized cluster analysis at the individual physician level. All data analyses were performed using Stata version 9.0 (Statacorp).