Currently Viewing:
The American Journal of Managed Care November 2013
Opioid Analgesic Treated Chronic Pain Patients at Risk for Problematic Use
Joseph Tkacz, MS; Jacqueline Pesa, PhD, MPH; Lien Vo, PharmD, MPH; Peter G. Kardel, MA; Hyong Un, MD; Joseph R. Volpicelli, MD, PhD; and Charles Ruetsch, PhD
Safety and Effectiveness of Mail Order Pharmacy Use in Diabetes
Julie A. Schmittdiel, PhD; Andrew J. Karter, PhD; Wendy T. Dyer, MS; James Chan, PharmD, PhD; and O. Kenrik Duru, MD, MSHS
Depression Self-Management Assistance Using Automated Telephonic Assessments and Social Support
John D. Piette, MSc, PhD; James E. Aikens, PhD; Ranak Trivedi, PhD; Diana Parrish, MSW; Connie Standiford, MD; Nicolle S. Marinec, MPH; Dana Striplin, MHSA; and Steven J. Bernstein, MD, MPH
Creating Peer Groups for Assessing and Comparing Nursing Home Performance
Margaret M. Byrne, PhD; Christina Daw, PhD; Ken Pietz, PhD; Brian Reis, BE; and Laura A. Petersen, MD, MPH
Upcoding Emergency Admissions for Non-Life-Threatening Injuries to Children
Zachary Pruitt, MHA; and Etienne Pracht, PhD
Currently Reading
Variations in the Service Quality of Medical Practices
Dan P. Ly, MD, MPP; and Sherry A. Glied, PhD
Collecting Mortality Data to Drive Real-Time Improvement in Suicide Prevention
Brian K. Ahmedani, PhD; M. Justin Coffey, MD; and C. Edward Coffey, MD

Variations in the Service Quality of Medical Practices

Dan P. Ly, MD, MPP; and Sherry A. Glied, PhD
Service quality (appointment lags and wait times) of primary care physician practices varies tremendously across the country and is associated with the organization of practices.
Objectives: To examine regional variation in the service quality of physician practices and to assess the association of this variation with the supply and organization of physicians.

Study Design: Secondary analyses of the Community Tracking Study (CTS) household and physician surveys.

Methods: A total of 40,339 individuals who had seen a primary care physician because of an illness or injury and 17,345 generalist physicians across 4 survey time periods in 60 CTS sites were included. Service quality measures used were lag between making an appointment and seeing a physician, and wait time at the physician’s office. Our supply measure was the physician-to-population ratio. Our organizational measure was the percentage of physicians in group practices. Multivariate regressions were performed to examine the relationship between service quality and the supply and organization of physicians.

Results: There was substantial variation in the service quality of physician visits across the country. For example, in 2003, the average wait time to see a doctor was 16 minutes in Milwaukee but more than 41 minutes in Miami; the average appointment lag for a sick visit in 2003 was 1.2 days in west-central Alabama but almost 6 days in Northwestern Washington. Service quality was not associated with the primary care physician-to-population ratio and had varying associations with the organization of practices.

Conclusions: Cross-site variation in service quality of care in primary care has been large, persistent, and associated with the organization of practices. Areas with higher primary care physician-to-population ratios had longer, not shorter, appointment lags.

Am J Manag Care. 2013;19(11):e378-e385
Cross-site variation in service quality (wait times and appointment lags) in primary care is large, persistent, and associated with the organization of practices.
  • This variation is robust to controls for patient characteristics.

  • Service quality is associated with the organization of practices.

  • Areas with higher primary care physician-to-population ratios had longer, not shorter, appointment lags.
The expansion of health insurance coverage to tens of millions of uninsured people has raised concerns about the adequacy of physician supply.1 An early report from Massachusetts, for example, suggested that waiting times to see primary care physicians had increased substantially after expansion of insurance coverage.2,3 Some research also suggested that the technical quality of medical care is higher in areas with more physicians,4 although this result was limited to generalist physicians.5

Yet there is little evidence on how the quality of the patient experience varies with physician supply. Later reports from Massachusetts suggested that reform had little impact on waiting times.6 Moreover, the relationship between physician supply and service quality may be tempered by how physicians are organized. Alternative organizations of practices may generate the same or better service quality with fewer physicians. 7-9 For example, service quality improvement tools (eg, electronic scheduling) may be more economical when physicians are organized into groups or participate in managed care organizations.

As the United States undergoes substantial changes to its healthcare system, including expansions in insurance coverage and changes to the delivery system, it is important to understand the landscape of service quality and to assess how that landscape may respond to changes in physician supply and organization. This study examines variations in 2 measures of service quality—waiting time and appointment lag—across the United States and assesses the association between these measures and the supply and organization of healthcare.


We drew data from the Community Tracking Study (CTS) Household Surveys (1996-1997, 1998-1999, 2000-2001, and 2003)10 and Community Tracking Study Physician Surveys (1996-1997, 1998-1999, 2000-2001, and 2004-2005)11 conducted by the Center for Studying Health System Change. Service quality measures used: 1) lag between making an appointment and seeing a physician and 2) wait time at the physician's office. These repeated cross-sectional surveys include representative samples of the population of households and physicians across 60 sites in the United States (51 metropolitan areas and 9 nonmetropolitan areas). The response rates for the CTS Household Surveys in chronological order were 65%, 63%, 61%, and 56.5%. The response rates for the CTS Physician Surveys in chronological order were 65.4%, 60.9%, 58.6%, and 52.4%. While the public-use data files for the CTS Household Survey contain CTS site identifiers, the public-use data files for the CTS Physician Survey do not, so we used the restricted-use data files for the latter survey. We matched the CTS Household Survey data and the CTS Physician Survey data by CTS site and year (the CTS Household Survey data from 2003 were matched with the CTS Physician Survey data from 2004-2005).

The CTS data have several limitations. As with the General Practice Assessment Questionnaire, Medical Expenditure Panel Survey, and Consumer Assessment of Healthcare Providers and Systems, measures of waiting time to see the doctor and time between obtaining an appointment and seeing a doctor were obtained by patient self-report and were not independently verified. Sample sizes for both the household and physician surveys in some of the smaller CTS sites were relatively small. The match between physician and household surveys was not exact in 2003 because no physician survey was conducted in 2003. While CTS did conduct a national survey for households in 2007 and 2010 and for physicians in 2008, those data are not available at the local level and are not suitable for these analyses. Notably, these survey years do not include wait times in the doctor’s office and, compared with prior survey years, do not include the exact number of days between appointment and doctor’s visit.

We focused on sick visits to primary care physicians (PCPs) by adult insured patients. We focused on sick visits because appointment times for return visits may be scheduled well in advance. We concentrated on primary care visits, as primary care has been the main area of concern about physician supply adequacy. Multivariate analyses controlled for patient age, sex, race, health status, insurance type (Medicare, Medicaid, private), education, employment status, marital status, and income. Health status was measured on a 4-point scale, and we dichotomized it to “fair or poor” (relative to good or better health). We included a measure of having at least 1 hospital stay in the previous year. Insurance was classified as Medicare (with or without supplemental coverage), Medicaid, or private insurance (employer based or individually purchased). Those with military insurance or other public coverage and the uninsured were excluded from the analyses to create a more uniform sample for cross-site comparisons. We defined education as less than high school, high school, some college, and college or more. We defined full-time employment as working 35 hours a week or more.

We focused on 2 measures of service quality: (1) the interval between making an appointment and seeing a physician (appointment lag) and (2) time in the physician’s office before being seen (wait time). Both measures refer to a patient’s last visit. We limited analyses of appointment lags to patients with lags of 21 or fewer days to capture only true sick care visits. (Analyses of  appointment lags longer than 21 days found that 84% of the lags occurred on 30-day intervals [30 days, 60 days, 90 days], suggesting that these were mainly scheduled follow-up visits. In analyses of patient satisfaction, the effect of appointment lag on satisfaction was negative and significant when lags were truncated at 21 days, but positive and significant without this truncation, again suggesting that long-scheduled visits were the main source of very long appointment lags.) This limitation excluded 12% of respondents; the basic results were not sensitive to the exclusion. We also repeated the analyses using alternative cutoffs (14 days and 30 days) and the log of the appointment lag. Results were very similar to those shown here.

Prior literature suggests that these dimensions of service quality do matter to patients.12-14 We verified the significance of these measures in our data by relating them to CTS respondents’ assessment of whether they were “very satisfied” with their healthcare on a 5-point Likert scale. In multivariate linear regressions, both measures were statistically significant predictors of satisfaction (Table 1).

Our primary measure of physician supply was the PCP-to-population ratio, which we constructed using PCP supply estimates from the annually published Physician Characteristics and Distribution in the U.S. and population estimates from the US Census.15 In sensitivity analyses, we also examined the relationship of service quality to 2 alternative measures of supply: (1) the PCP-to-insured population ratio, constructed as above but adjusting the denominator to reflect the proportion of the population in the CTS household sample holding any type of health insurance, and (2) the percentage of physicians at a CTS site who reported that they accepted all new Medicare patients (implying that there are enough physicians in the area for physicians to be willing to accept additional patients at Medicare reimbursement levels). Since Physician Characteristics and Distribution in the U.S. did not have physician supply estimates for nonmetropolitan CTS sites, we excluded the 9 nonmetropolitan CTS sites in analyses using physician-to-population ratios.

One potential concern about the use of contemporaneous measures of physician supply in analyses of appointment lags and wait times is that the supply of physicians may respond to wait times and appointment lags. To assess the importance of this potential reverse causality, we calculated Pearson correlation coefficients between wait times and appointment lags in 1996 and the subsequent change in physician supply from 1996 to 2000 (and from 1996 to 2005).

The primary measure of physician organization used was the percentage of physicians in a CTS site in group practices of more than 10 physicians (group practices with 7 or more physicians are in the 75th percentile for the number of physicians in a practice). To separate the effects of organization from those of payment, we included in the regressions the average share of revenue in a CTS site from capitation. In our analyses we also controlled for the share of physicians employed by institutions such as hospitals or medical schools, as these organizations may have different incentives than do private practitioners.

To assess the extent of variation in service quality across the country, we computed patient risk-adjusted wait times and appointment lags by estimating average residuals by CTS site and year of each service quality measure after controlling for the patient sociodemographic characteristics listed above. We ordered the residuals by quartiles and mapped them. To assess the persistence of site-specific variation in wait times and appointment lags across survey years, we calculated Spearman’s rank correlation coefficients of the quartiles of these residuals.

Next, we calculated weighted means of the supply and organizational measures from the physician survey and matched these by year and CTS site to the household survey. Following the method of Fisher and colleagues, these variables were measured at the site-year level, which is appropriate as they reflected market-level conditions.16,17 In multivariate analyses, we examined the association between these CTS site-level supply and organizational variables and service quality.

We analyzed physician data in SUDAAN to adjust for the complex design of the CTS survey and for clustering at the CTS site level.18 Household data were analyzed using the complex survey modules in Stata release 10.0 to take advantage of statistical procedures in Stata not found in SUDAAN.19 These complex survey modules accounted for the clustering of observations at the site level and the repeated sampling of some individuals. For site-specific estimates of the household survey, the variance estimates from Stata were identical to those generated by SUDAAN.20 In multivariate analyses, we pooled data across all 4 survey years. In all analyses we included year dummies to capture time-varying effects, and we used robust standard errors to adjust for heterogeneity. We therefore ran regressions of the form:

Q = b0 + b1S + b2O + b3X + b4T + εi

where Q is one of the above measures of service quality, S is one of the above measures of physician supply, O is one of the above measures of physician organization, X is a vector of patient characteristics, T represents year dummies, and ε is the error term (taking into account the complex survey design).


Copyright AJMC 2006-2019 Clinical Care Targeted Communications Group, LLC. All Rights Reserved.
Welcome the the new and improved, the premier managed market network. Tell us about yourself so that we can serve you better.
Sign Up