Specifically, this study of Washington State addresses the following questions:

• What is the total contribution to generalist care made by NPCs?
• What is the role of NPCs in providing generalist care in rural Health Professional Shortage Areas (HPSAs)?
• What proportion of the total generalist care provided by women is provided by women NPCs?

An important issue in health workforce analysis is how to count the contribution to patient care of each provider and each provider type. Simple head counts of providers are unlikely to produce realistic estimates of the actual supply of health care available to a population (Larson, Ballweg, and Hart 2001; Ricketts, Hart, and Pirani 2000). Differences in training, location, specialty, in-patient care activities, experience, scope of practice, and full-time/part-time status create large differences in the number of visits that a given clinician is likely to perform during a week. Data from the American Medical Association (Randolph, Seidman, and Pasko 1997), for example, indicate that an average family physician provides 105 ambulatory patient visits each week, a general pediatrician 95, and a general internist about 65. If estimates of available care are to include NPCs, basing estimates of available care on head counts becomes yet more doubtful because so little is known about the productivity of NPCs and their total contribution to care (Scheffler, Waitzman, and Hillman 1996). In this study, we use state licensing data and self-reports of outpatient visit productivity to count contribution to care by providers in family physician full-time equivalents (FTEs). By measuring productivity in this way, we can correct for individual differences in productivity and directly compare productivity across professions.

Because the analysis is based on individual-level productivity data, we are able to base our analysis on FTEs rather than head counts of providers. While many states and other jurisdictions do not have the data necessary to carry out such an analysis, this work presents some empirically derived “productivity adjustment factors” for different provider types in different settings. If we can determine, for example, that full-time PAs perform 70 percent of the number of visits performed by full-time family physicians, we can, in the absence of actual visit data, use the .7 productivity adjustment to estimate the contribution to care made by a PA in meeting the visit requirements of a population. Productivity adjustments can be used to estimate available care based on provider type, specialty, and location when productivity data are not available.

Data on physicians who did not respond to the state licensure survey were obtained by linking American Medical Association (AMA) Masterfile data to the licensing data, reducing the effective rate of missing specialty data for physicians to only 1.7 percent. Though the AMA data provided specialty information, it did not include data on patient visits.

Where: t=count of known active generalists

g=proportion of active generalists among known providers

s=count of unknown providers

r=rural/urban location

Imputation of missing specialty data in this manner is a conservative measure that guards against undercounting the number of providers of any given type in a population.

Counts of number of outpatient visits performed each week from providers who completed the survey were used to estimate the number of generalist visits performed by providers who did not provide specialty and/or productivity information (equation (2)):

Where: B=total visits

V=reported visits

t=provider type (FP, GIM, Ped, PA, NP)

M=median visits from known providers

r=rural or urban

a=proportion of providers actively practicing

p=proportion of provider type in primary care

(this will be 1 for known FPs, GIMs, and Peds who did not have productivity data)

The productivity adjustment factors and median outpatient visits (M and p) applied to all unknown providers are reported below. All were derived from the provider population with complete survey data.

To impute missing items for providers who responded to the survey we systematically worked our way through all the large and small groups of providers over a six-month period. The imputation decisions usually involved either imputing missing productivity data (for clearly identified generalists only) or deciding whether or not to count the visits of particular combinations of provider type and primary and secondary specialties (e.g., family medicine, community medicine, and public health) as generalists. The research team met weekly to discuss the various options and review analyses (including regressions), and used our workforce experience and the data to create consistent decision rules. Many of the analyses involved printing out all members of a small subgroup and examining the data by hand per all the relevant reported items and combinations of items. For example, if a physician listed his or her first specialty type as community health (or as a nonclinical position, public health, etc.) and listed FP as a second specialty and provided information on direct patient care and visits, they were included as generalists. Many other combinations were not included (e.g., a physician who had a nonclinical position and family physician combination, but did not report patient visits was not included as contributing generalist visits). In selected cases where visits were missing but direct patient care hours was present, visits were estimated. Both the AMA Masterfile and AHEC data were integrated into this process.

In the case of providers who did not respond to the survey, we generally assumed that nonrespondents were similar to respondents. As indicated above, regression analysis results were suspect, given the limited license variables and its poor performance in tests. Therefore, we created algorithms that attributed the characteristics of the respondents to the nonrespondents using available data: license data, AMA data for physicians, and AHEC data (e.g., if the AHECs reported that a nonrespondent was practicing full-time as a generalist at a site, we reflected this in the database). We then extrapolated the available information to the nonrespondents. After all steps to identify specialty via license data, AMA data, and AHEC information, we had to impute specialty for only 8 percent of the 15,606 active specialist and generalist providers in the state (Hart and Palazzo [in press]). The use of information from the AHECs reduced the possibility of error in estimating care in the low-population HSAs where the consequences of an error would be the most severe.

In some cases, we chose to exclude providers rather than impute specialty or productivity data. For example, providers who did not provide direct patient hours or visits (either as missing questionnaire items or through survey nonresponse) and who were age 70 or older, were excluded from the analyses. It would have introduced more error into the estimates to impute to all licensed 70 or older providers (there being many), when few would have produced meaningful numbers of visits. Instead we chose to be conservative and only include those who also reported their actual visits. In this case, the assumption was that the nonrespondents were much more likely to not be practicing than the respondents. The rule of thumb was one of reducing the largest source of probable error. An exhaustive description of the imputation and exclusion methods used in this study can be found in Hart and Palazzo (in press).

It should be emphasized that the conversion to family physician FTEs is done only for convenience and the facilitation of comparisons. The proportion of care provided by NPs, PAs, and physicians can be calculated directly from the components of equation (2). These proportions remain the same regardless of whether they are converted to family physician FTEs, or to any other provider standard FTE; one could, for example, decide to estimate FTEs based on median generalist productivity. In this study, available primary care FTEs in a given area are calculated (equation(3)):

F=total family physician FTEs available

B=total visits (from equations (2))

t=provider type (FP, GIM, Ped, PA, NP)

For example, if a family physician in a given HSA provides 110 visits each week and an NP and a PA provide 60 and 40 visits (respectively), then the provider supply in the HSA is estimated at 210 visits or two family physician FTEs.

Table 1 Active Generalist Providers in Washington State, 1998–1999

Table 2 Factors Used in Estimating the Productivity of Providers Who Were Missing Productivity Data

Table 3 Active Identified and Imputed Generalist Providers by Rural/Urban Location in Washington State, 1998–1999

Figure 1 Generalist Head Counts and FTEs, by Provider Type, Washington State, 1998–1999

There are some differences in the contribution to care made by NPCs in rural versus urban areas of Washington. Overall, NPCs provide 24.7 percent of the total generalist outpatient visits in rural areas (10.3 percent by NPs, 14.4 percent by PAs) compared to 20.1 percent in urban areas (p=.014). In both rural and urban settings, NPs provide about 10 percent of the outpatient visits; the rural/urban difference in NPC contribution is primarily attributable to the larger proportion of total visits provided by rural PAs. The range of NPC contribution to generalist care in rural HSAs is quite wide. In five rural HSAs, NPCs provide less than 10 percent of total generalist visits. In three HSAs, NPCs provide over 75 percent of visits. In the majority of rural HSAs, NPCs provide between 20 and 30 percent of the generalist outpatient visits.

Table 4 Generalist Provider Supply (in FTEs) in Rural Geographic HPSAs, Washington State, 1989–1999

Table 5 Percentage of Female Generalist Providers by Provider Type (in FTEs) by Rural/Urban Location, Washington State, 1998–1999

The contribution of NPCs to generalist care in rural geographic HPSAs appeared to be slightly higher than in non-HPSAs, but the observed differences were not found to be statistically significant. We had hypothesized that NPCs would make up a larger part of the care system in HPSAs than in non-HPSAs because NPCs generally cost less and are believed to be easier to recruit and retain than physicians. The data presented here do not generally support that hypothesis. The PAs do appear to make a larger contribution in HSAs with large HPSA populations, compared to non–HPSA HSAs.

Though women make up an increasing part of the generalist physician workforce and are the providers of choice of many female patients (Fennema, Meyer, and Owen 1990), rural medicine has been relatively unattractive to women physicians (Doescher, Ellsbury, and Hart 2000). We hypothesized, therefore, that in rural settings, NPCs would provide a larger share of the care provided by women. This hypothesis was sustained. In rural settings female physicians provided less than half (49.3 percent) of the FTEs provided by women. In contrast, women physicians provided 63.5 percent of the generalist care provided by women in urban settings. In both rural and urban settings, NPs provided the majority of female NPC FTEs.

Because we wanted to assess the total contribution to generalist care made by physicians, NPs, and PAs, we could not simply exclude cases missing productivity data. It was necessary to develop the imputation method described above for estimating productivity (and sometimes specialty and active/nonactive status) based on the known population of providers. Any productivity estimate is bound to involve some inaccuracy. The degree of error will vary with the proportion of cases being estimated compared to the proportion that is known and the degree to which the respondents were representative of the nonrespondents. This is most problematic for the NP estimates since NPs responded to the survey instrument at a much lower rate than physicians or PAs. For example, if survey nonrespondent NPs were less likely to be actively practicing, their contribution to primary care would be less than estimated.

In interpreting the results of the study the reader is cautioned that observed differences in productivity across professions may reflect either differences in hourly productivity or differences in number of hours worked (or, most likely, both) (Larson, Ballweg, and Hart 2001). Since our purpose here was to assess the overall contribution of NPCs to generalist ambulatory care, we did not adjust our contribution estimates for hours worked. Our estimates are based on reports of total ambulatory visits performed weekly.

Any time physicians are compared to other clinicians the question of substitutability is raised. Washington's less restrictive practice environment for NPs and PAs may allow for a level of substitution not found in other states where prescriptive authority, insurance requirements, and supervisory arrangements may allow considerably less substitution. Caution must also be exercised in employing these findings in discussions of provider shortage mitigation. The NPCs may be providing 25 percent of care in a location and the overall provider population ratio may seem reasonable—but this does not mean, necessarily, that there is no physician shortage in that place. A population with a high disease burden of severe acute or chronic illness, for example, may require a higher number of visits to physicians (as opposed to NPCs) than a population with a lower burden. The health of a local population, patient preferences, facilities, and a host of other factors make up the background that determines the constellation of services and providers that will effectively address the health care needs of a given population. Access to primary care ambulatory visits is important, but hospitals, emergency care systems, and public health systems all play vital roles in addressing those needs as well. Finally, the FTE estimates shown above should be treated with caution. While the proportion-of-care estimates are stable, the FTE estimates shown are based on national (and Washington state) estimates of full-time family physician productivity (105 outpatient visits per week). In some situations, another FTE standard, such as median generalist productivity in a given state, might be deemed more appropriate.

The results of this study point to the need for empirically based estimates of provider counts and provider productivity. For example, proposed changes to HPSA designation rules include a change that would count PAs and NPs as .5 of a primary care physician FTE (Department of Health and Human Services 1998). Under the proposed rules, all primary care physicians are counted as one FTE. No adjustments are made for specialty, part-time status, or any actual known productivity data. The results of this study indicate strongly that simple provider head counts are likely to result in severe overestimates of available care. Recall, for example, that we estimated that a total of 4,145 generalist physicians produce only 2,760 family physician FTEs. The NP and PA counts were also decreased substantially when turned into FTEs.

While head count to FTE adjustments are important, it is also important to consider productivity differences across and within provider groups. Work based on a national sample of PAs (Larson, Ballweg, and Hart 2001), as well as some earlier work (Cyr 1985; Scheffler, Waitzman, and Hillman 1996), for example, suggests that generalist PAs are likely to perform about 75 percent as many outpatient visits as generalist physicians, not the 50 percent suggested in the proposed HPSA designation rules. Nurse practitioners appear more likely to work part-time than other providers. However, their hourly visit productivity appears to be only slightly lower than for PAs. In a case where one is trying to estimate the available care in a given area, it is obviously important to know the part-time/full-time distribution of providers both within and across professions. Taken together, the analyses presented above highlight the inherent and severe weaknesses associated with using provider head count to population ratios as measures of available care. More accurate estimates of provider productivity and contribution-to-care can significantly enhance our ability to assess provider shortage, identify solutions, and plan for the education of future health care providers. Such estimates can be made using the methods outlined above with data that are relatively simple and inexpensive to collect.

This study describes the overall configuration of generalist care in Washington State in 1998–1999 and shows that NPCs provide over 20 percent of the generalist ambulatory visits performed in urban areas and 24 percent in rural areas. Additionally, rural areas of Washington State are particularly dependent on NPCs for women primary care providers. Clearly, NPC contributions must be taken into account whenever one is trying to assess the availability of generalist care. At the same time, care must be taken with the methods used to count providers and estimate productivity. When working from provider head count data, the use of productivity adjustment procedures, such as those described above, are essential for accurate estimates of provider shortages and available care.

The authors would like to thank the Eastern Washington Area Health Education Center, the Western Washington Area Health Education Center, the Washington State Department of Health, and Vince Schueler of the Washington State Office of Community and Rural Health for their assistance with data collection.