Holmes, George M. PhD; Morrison, Marisa; Pathman, Donald E. MD, MPH; Fraher, Erin PhD, MPP
A common approach in physician workforce modeling and policy analysis is to assess whether there is a physician shortage by considering each individual specialty to be distinct, defined by the different training experienced by and unique scope of services provided by its practitioners.1–4 This “siloed” conception of specialties ignores the reality that the scope of medical services that physicians of different specialties provide often overlaps. This traditional approach also treats all physicians within a single specialty as identical and therefore interchangeable, even though individuals within a given specialty offer different mixes of services because of their particular training and interests.
An alternative health care workforce modeling approach exists. (In this article, we refer to “physicians” for expositional simplicity, although the model could easily be extended to other clinicians such as physician assistants and advanced practice nurses. We use “providers” or “workforce” to refer to this broader group.) The approach we describe herein acknowledges that physicians of different specialties may overlap in the scope of services they provide, and more realistically recognizes that a community can increase access to (for example) diabetes care by increasing its supply of family physicians, internists, endocrinologists, or some mix of these (or other) groups. This alternative allows for multiple combinations of physician specialties to provide a specified group of medical services but still recognizes that certain specialties are more likely to provide certain types of health care services.
Heterogeneity in the services provided within a specialty also characterizes physician practice. For instance, some internists devote a greater proportion of their visits to respiratory conditions, whereas others focus more on circulatory conditions. Few researchers have conducted scholarly work exploring either within-specialty heterogeneity or between-specialty service overlap, despite the importance of these realities to the solutions that could flow from physician workforce models. We suggest that these related concepts represent two facets of physician plasticity. This article’s objective is to describe the concept of plasticity and to explain how we have applied plasticity in a model that projects the sufficiency of physician supply to meet a specific population’s expected health care utilization.
We define plasticity as the notion that physicians of different specialties may offer similar medical services to treat or manage a condition (between-specialty plasticity), and that individual physicians within a specific specialty may differ in the services they provide (within-specialty plasticity). Within-specialty plasticity and between-specialty plasticity are related: Within a given community, the specific scope and relative mix of services provided by individuals within a single specialty can affect patterns of service provided by physicians in other specialties. Plasticity allows policy makers, investigators, or others the ability to measure not only the mix and supply of physicians in different specialties within a community but also the population’s utilization of health care services, and in turn, the scope and mix of services that different combinations of physicians in different specialties can provide to the community.
Within-specialty plasticity is, as mentioned above, the concept that individual physicians within the same specialty may each provide a different mix and scope of services. Evidence indicates that practice patterns among physicians within the same specialty differ according to physician demographic characteristics (e.g., age and gender) as well as local conditions (e.g., geographic region or rural or urban location).5–10 This existing literature on practice patterns has generally focused on differences in scopes of services provided by family physicians and general surgeons,5–10 although other studies have noted differences in the mix of services provided by rural and urban physicians in obstetrics–gynecology and in pediatrics.11,12
This literature tends to define the scope of physician practice according to the relative volume and mix of procedures and services that physicians perform.5,9,13 In contrast, we define the scope and mix of physician service provision according to the relative number or proportion of patient visits across categories of clinical conditions—such as digestive diseases or skin diseases—that physicians address during all patient visits in a given time frame (e.g., a month or year). We call these categories of health care services clinical service areas (CSAs), and we based these categories on the Agency for Healthcare Research and Quality’s (AHRQ’s) clinical classification categories.14 We expanded AHRQ’s clinical classifications by creating an additional CSA for preventive care.
Analyses of freely available, deidentified 2008 National Ambulatory Medical Care Survey data confirm that variation exists in terms of the percentage of patient visits dedicated to each CSA among physicians within a single specialty. These analyses also demonstrate how the degree of within-specialty plasticity varies by specialty. Figures 1A and 1B show the heterogeneity of services provided by random samples of, respectively, 10 internists and 10 dermatologists. Figure 1A shows that the distribution of visits by CSA varies widely across individual internal medicine physicians—particularly for the respiratory system diseases, circulatory system diseases, and endocrine disorders. This wide variation suggests that internists exhibit a high degree of within-specialty plasticity. In addition, the figure shows that internists have a generally broad scope of services, a hallmark of a primary care specialty.
Figure 1B shows that within-specialty plasticity among dermatologists is limited; care is concentrated in the skin disease and cancer CSAs. Furthermore, Figure 1B demonstrates that dermatologists provide care within fewer CSAs than do internists.
Between-specialty plasticity—the variance and overlap of services provided across specialties—is not well documented in the literature. Pathman and Tropman15 (1995) describe how the supply of obstetricians in a rural county correlates with whether family physicians in that county offer maternity care. Pathman and Ricketts16 (2009) document how rural general surgeon service provision depends on the referral decisions and content of practice of rural family physicians (and vice versa).
Analyses of 2009 Medical Expenditure Panel Survey (MEPS) data confirm that the scope of service of physicians from different specialties overlaps. Table 1 shows, for example, how five specialties, each with a broad scope of services, have divided patient visits among four selected CSAs. (We present in this table just a subset of all the specialties and CSAs we consider so that readers can interpret the data more easily.)
An important step in assessing the “sufficiency of physician supply” is gauging the extent to which between-specialty plasticity is possible—that is, whether the existing physicians in different specialties within a community have the plasticity to divide their time between different types of health services to meet the community’s health care utilization patterns. For example, in communities lacking an adequate supply of obstetrician–gynecologists, family physicians take on labor and delivery services.15 Theoretically, when family physicians provide more prenatal care, physicians of other specialties could increase their number of visits for services that family physicians previously provided: that is, internists and cardiologists could increase their circulatory visits, and psychiatrists (and psychologists) could provide more mental health visits. This process allows visits to be redistributed among specialties so as to begin to achieve a new balance reflecting physician supply and community health services utilization.
Applying Plasticity in Workforce Modeling
The plasticity element discussed herein is one aspect of a broader model (described in greater detail elsewhere17) to project physician supply and health care utilization. In that model, plasticity (both within- and between-specialty) provides the mechanism for determining the sufficiency of physician supply to meet forecasted physician visit utilization by CSA for a specific geographic region. The model projects physician supply and physician visit utilization before reconciling the two using a plasticity matrix. The model uses American Medical Association (AMA) Masterfile data, American Board of Medical Specialty certification data, and resident supply data to determine physician supply, and it uses MEPS data and community-level demographic and economic information to generate community-level forecasts of physician visit utilization for each CSA.
The plasticity matrix maps likely patient visits within CSAs to likely available physicians. The plasticity matrix describes two elements critical to determining the adequacy of physician supply within a community: first, how visits by CSA are distributed on average between physicians in different specialties; and second, how physicians of a single specialty, on average, divide their patient visits across CSAs. Importantly, we reduced the AMA Masterfile’s list of 243 primary specialties to 37 specialty categories (henceforth, simply “specialty”) for reasons of feasibility. We used data from the 2009 MEPS to determine the percentage of each specialty’s visits to dedicate to each CSA and to populate the plasticity matrix’s cells. In the current version of our physician supply model,17 we have 18 CSAs and 37 specialties.
Table 1 provides a simplified version of this matrix with only five physician specialties and four CSAs. In this simplified model, cardiologists provide 34% of all patient visits for circulatory conditions in a single year (in this case, 2009), and a small percentage (< 4%) combined for patient visits related to neoplasms, respiratory illnesses, and pregnancy/childbirth. Family physicians provide 38% of circulatory visits but also provide the majority of visits for respiratory conditions compared with cardiologists, dermatologists, obstetricians–gynecologists, and internists. In other words, among a subset of CSAs—specifically, neoplasms, circulatory-related conditions, respiratory conditions, and pregnancy/childbirth visits—cardiologists provide care for essentially only circulatory visits, whereas family physicians provide the bulk of their visits in circulatory and respiratory conditions. Thus, family physicians are more plastic—that is, they can provide visits for two of the illustrated conditions, whereas cardiologists focus on only one CSA.
The simplified plasticity matrix in Table 1 represents the average number of physicians and the national average number of visits by CSA in 2011. As explained below, the distribution of visits for CSA between specialties in a specific community is determined iteratively, and the final estimate can be used to assess the sufficiency of physician supply for a specific community. The algorithm for adjusting the service distribution accounts both for local conditions, including the supply and specialty mix of physicians in the community, and for some bounds on service provision by specialty (e.g., psychiatrists cannot provide circulatory visits).
Here we demonstrate how to implement plasticity in determining the sufficiency of physician supply to meet estimated physician visit utilization in a single county—Wayne County, North Carolina. For clarity, we describe fewer CSAs and specialties than are used in our full model. Likewise, in this example we do not discuss the specifics of our full model that are outside the plasticity element (e.g., projecting health care use at a community level and aggregating physicians into distinct specialties).
Step 1. Model the health care utilization of the community (e.g., county) as a function of local conditions
We first estimate Wayne County’s expected generation of physician visits (utilization) by CSA in 2011 as a function of the 2009 MEPS data, adjusted for the county’s age-sex-race population counts, insurance coverage, health risk factors, and income as of 2011. The algorithm generating these estimates is outside the scope of this paper (contact authors for further information, if interested), and of course, local circumstances may lead to utilization that differs from the estimates presented here. We estimate that Wayne County’s approximately 114,000 residents would generate 389,867 annual office visits for the 18 CSAs (Table 2).
Step 2: Collect data on local physician supply
In 2011, Wayne County had 3 cardiologists, 29 family physicians, 18 internists, and multiple other physicians (excluded from this simplified example). Again, the specific process used to develop these counts is outside the scope of this application (contact the authors for further information).
Step 3: Initial allocation of visits
At this point in the process, we have generated forecasts both of county-level utilization by CSA and of physician supply by specialty. Because the 18 CSAs for which the population is seeking care in Wayne County (Table 2) could be managed by a variety of configurations of physicians within the 37 specialties, the physician projection model uses the plasticity matrix (simplified version in Table 1) to distribute visits by CSA across specialties according to the typical, national-level distribution. If the care-seeking patterns in Wayne County are similar to those nationally, approximately 34% of circulatory visits should go to cardiologists, 38% should go to family physicians, and 26% should go to internists.
The model initially allocates the total physician visits for Wayne County proportionally across physicians of various specialty categories according to these national averages. Calculating that 34.4% (a more precise figure; Table 1) of Wayne County’s 48,585 circulatory visits (Table 2) go to cardiologists, we expect 16,713 visits to cardiologists (i.e., 48,585 × 0.344). However, we have to consider the volume of services that Wayne County’s three cardiologists can supply. Nationally, cardiologists provide approximately 95.9% (23.4 million circulatory visits / 24.4 million total visits) of their visits in the circulatory CSA. Assuming that physicians provide 2,500 patient visits per year18 (the number of visits used here is simply for illustration purposes; alternatively, the model could use benchmarked data if available by specialty and practice type [e.g., from the Medical Group Management Association]), then 95.9% × 2,500 visits × 3 cardiologists = 7,192.5 visits—fewer than half of the 16,713 visits that would be expected by national patterns to be seen by cardiologists.
Therefore, after this first allocation and given national visit patterns, the model indicates that cardiologists in Wayne County can expect approximately 9,520.5 (i.e., 16,713 − 7,192.5) patient visits for circulatory conditions beyond the county’s existing cardiologist capacity. Meanwhile, at the initial allocation, Wayne County’s internists have “excess capacity” of about 7,700 visits—only about 80% of their visit capacity is allocated (not shown). Thus, there is an imbalance: Utilization of cardiologist visits for circulatory visits outstrips supply, but internists have excess capacity. In other words, there are patients who would “typically” go to a cardiologist for their circulatory visit but cannot get in, and internists have unfilled appointments.
Step 4: Adjusting physicians’ scope of services
The model assumes that physicians adjust their service portfolio away from the national average to account for the local population’s unmet health services utilization; that is, physicians “capture” unmet visits within the bounds of their scope of service provision. In the Wayne County case, because internists have 20% excess capacity, they can “capture” some of the circulatory visits that are used in the community but exceed the capacity under the national allocation. In this case, after the model iteratively allocates circulatory visits, optimizing the mix over multiple passes in order to identify the distribution that minimizes both overcapacity and undercapacity across CSAs, 5,891 of the circulatory visits that could not be provided by the cardiologists will be allocated to internists (in addition to the 12,632 initially allocated). This reduces but does not eliminate internists’ excess capacity; this step is repeated to identify specialties with excess capacity that can expand into CSAs that have unmet need.
After multiple iterations, the model reveals the distribution of services that minimizes excess capacity, aligns with bounded service distribution by specialty, and accounts for Wayne County’s physician supply by specialty and projected utilization of physician visits by CSA. Incidentally, the final, “equilibrated” distribution for Wayne County shows that its internists provide relatively more circulatory visits and fewer respiratory visits than the national average for internists.
Step 5: Assess sufficiency
The completed projected distribution of CSAs to physician specialties allows for comparing the capacity for patient visits for a particular CSA with the predicted utilization of visits for that CSA. If the capacity is less than the predicted utilization, there is a shortage of physicians for that service. In Wayne County, after the algorithm model iterates across all CSAs and specialty types, 3% of circulatory visits cannot be filled by the county’s physicians. Other services (e.g., respiratory, pregnancy/childbirth) are completely met; others (e.g., cancer, mental health) have considerable unmet utilization.
Plasticity can be used to simulate how policy changes or other factors influencing physician supply and health care utilization will affect the sufficiency of supply at various geographic levels. For example, if policies led to an increase in the number of practicing cardiologists in Wayne County, the cardiologist supply would be sufficient to support the national cross-specialty distribution of cardiology visits, affecting how the model distributes visits by CSA across specialty categories at the local level in Wayne County. This net effect of such a hypothetical increased number of cardiologists in Wayne County is that family physicians would spend 6.4% of their visits caring for circulatory conditions rather than the national average of 24% (using the full plasticity matrix, not shown). Thus, the model predicts that an increased cardiologist supply in Wayne County would lead local family physicians to decrease the number of patient visits for a service that is relatively well met (circulatory) and focus, instead, on service provision for CSAs with unmet utilization (e.g., musculoskeletal, in this case). Likewise, changes to factors that affect utilization conditions, such as in the number of insured patients in a community, can be modeled.
The proposed approach is novel in that it acknowledges real-world variability in the scope of service of physicians in different specialties. Although we focus on two specific aspects—within-specialty and between-specialty plasticity—other dimensions could be considered. For example, scope may vary geographically irrespective of the local relative supply and demand conditions, potentially driven by either local practice and referral patterns, characteristics of physicians (e.g., those with broader versus more narrow training, those who are working part-time or who are semiretired), patient characteristics (e.g., mobility and willingness to travel to access care), technology (e.g., the availability of telemedicine), and state policies (e.g., licensure regulations, malpractice rates).
There are several limitations to the concept of plasticity as we have described it that suggest avenues for further research and modeling efforts. First, we have discussed plasticity only in the context of physicians; the scope of services that nonphysician clinicians provide overlaps with physicians’ scope of services.19,20 Future efforts to incorporate plasticity into workforce modeling should address this interprofessional plasticity in service provision.
Moreover, additional research is needed to identify the factors that shape how physicians apportion their time to meet a community’s health care needs. Presumably, the factors that influence the scope of services an individual physician provides include economic forces (e.g., competition as well as opportunities to partner with other clinicians, patient demand, and service reimbursement levels), clinicians’ training backgrounds, the physician’s current skills and practice preferences, and his or her professionally instilled sense of responsibility to patients. To the extent that the data underlying the model described here—local provider supply, community-level utilization, and the national “plasticity matrix” used as a baseline—are inaccurate or inadequate, the model will give inaccurate results. Further, the model does not consider differences in outcomes or quality or patient satisfaction, which may vary considerably across specialties. The use of the “visit” as the unit of output means that the model provides better estimates for outpatient than inpatient settings where utilization is not measured in visits and where data on care provided by some specialties (e.g., pathologists and radiologists) are difficult to access.
Although this approach extends current workforce science, like all models it is based on simplifying assumptions. As mentioned, the AMA Masterfile lists 243 primary specialties, and modeling each specialty separately is empirically infeasible, so analysts must “group” specialties into categories which differ in “typical” scope of services (e.g., interventional cardiologist versus an office-based general cardiologist). On the utilization side, a core assumption is visit homogeneity, but visits will vary within CSA (e.g., chronic renal failure and endometriosis are both classified as “geritourinary conditions”) and by complexity (initial visit versus follow-up; number of chronic conditions, patient age). Furthermore, the same visit may take longer if seen by one specialist than another. Applying the approach outlined above is a technical challenge, and the additional complexity may not sufficiently improve the predictive power of the model.
A future direction would be to populate the model with better specialty-specific data on the profile of services those specialists provide (e.g., from the American Board of Medical Specialties maintenance of certification files); current data are limited (e.g., by specific payer). An additional question to consider is whether providers in “surplus” or “shortage” areas behave differently; for example, do they select “more attractive” (healthier, better reimbursed) patients in shortage areas or induce demand in areas with surplus? This provider feedback to the model could be useful in explaining observed cost and practice pattern differences.21
In this article, we have suggested that the concept of plasticity—the notion that the scope of services provided by physicians of different specialties overlaps and that the mix and scope of services provided by physicians within a given specialty is heterogeneous—could make an important contribution to workforce modeling, research, and policy analysis. This alternative modeling framework attempts to incorporate more of a real-world approach to how care is apportioned within communities, moving beyond treating specialties as silos. It also takes a population-based approach to workforce planning by first identifying the population’s predicted utilization of health care services and then identifying a configuration of physician specialties that could meet the predicted utilization and minimize unmet need. As additional literature and data emerge on the factors affecting the scope and distribution of services provided by clinicians of different specialties and professions, this approach could be more finely calibrated and expanded to include other providers so that the effect of task-shifting and teamwork in new models of care could be incorporated.
Acknowledgments: The authors wish to thank Andy Knapton, MSc, and Thomas Ricketts, PhD, and participants at the 2012 Association of American Medical Colleges Workforce Conference and the 2012 Academy Health Annual Research Meeting who provided helpful comments on the proposed approach.
3. Erikson C, Salsberg E, Forte G, Bruinooge S, Goldstein M. Future supply and demand for oncologists: Challenges to assuring access to oncology services. J Oncol Pract. 2007;3:79–86
4. Grover A, Gorman K, Dall TM, et al. Shortage of cardiothoracic surgeons is likely by 2020. Circulation. 2009;120:488–494
5. Chan BT. The declining comprehensiveness of primary care. CMAJ. 2002;166:429–434
6. Chaytors RG, Szafran O, Crutcher RA. Rural–urban and gender differences in procedures performed by family practice residency graduates. Fam Med. 2001;33:766–771
7. Harris JD, Hosford CC, Sticca RP. A comprehensive analysis of surgical procedures in rural surgery practices. Am J Surg. 2010;200:820–825
8. Hutten-Czapski P, Pitblado R, Slade S. Short report: Scope of family practice in rural and urban settings. Can Fam Physician. 2004;50:1548–1550
9. King J, Fraher EP, Ricketts TC, Charles A, Sheldon GF, Meyer AA. Characteristics of practice among rural and urban general surgeons in North Carolina. Ann Surg. 2009;249:1052–1060
10. Valentine RJ, Jones A, Biester TW, Cogbill TH, Borman KR, Rhodes RS. General surgery workloads and practice patterns in the United States, 2007 to 2009: A 10-year update from the American Board of Surgery. Ann Surg. 2011;254:520–525
11. Baldwin LM, Rosenblatt RA, Schneeweiss R, Lishner DM, Hart LG. Rural and urban physicians: Does the content of their Medicare practices differ? J Rural Health. 1999;15:240–251
12. Feigin RD, Drutz JE, Smith EO, Collins CR. Practice variations by population: Training significance. Pediatrics. 1996;98(2 pt 1):186–190
13. Wong E, Stewart M. Predicting the scope of practice of family physicians. Can Fam Physician. 2010;56:e219–e225
14. Agency for Healthcare Research and Quality. . Healthcare Cost and Utilization Project (HCUP). Clinical Classifications Software (CCS) for ICD-9-CM. August 2013 Rockville, Md Agency for Healthcare Research and Quality www.hcup-us.ahrq.gov/toolssoftware/ccs/ccs.jsp
. Accessed August 30, 2013
15. Pathman D, Tropman S. Obstetrical practice among new rural family physicians. J Fam Pract. 1995;40:457–464
16. Pathman DE, Ricketts TC 3rd. Interdependence of general surgeons and primary care physicians in rural communities. Surg Clin North Am. 2009;89:1293–1302, vii
18. Yarnall KS, Pollak KI, Østbye T, Krause KM, Michener JL. Primary care: Is there enough time for prevention? Am J Public Health. 2003;93:635–641
19. Hooker RS, Everett CM. The contributions of physician assistants in primary care systems. Health Soc Care Community. 2012;20:20–31
20. Laurant M, Reeves D, Hermens R, Braspenning J, Grol R, Sibbald B. Substitution of doctors by nurses in primary care. Cochrane Database Syst Rev. 2005(2):CD001271
21. Welch WP, Miller ME, Welch HG, Fisher ES, Wennberg JE. Geographic variation in expenditures for physicians’ services in the United States. N Engl J Med. 1993;328:621–627