The cost-effectiveness of physician assistants/associates: A systematic review of international evidence

Background The global utilization of the physician assistant/associate (PA) is growing. Their increasing presence is in response to the rising demands of demographic changes, new developments in healthcare, and physician shortages. While PAs are present on four continents, the evidence of whether their employment contributes to more efficient healthcare has not been assessed in the aggregate. We undertook a systematic review of the literature on PA cost-effectiveness as compared to physicians. Cost-effectiveness was operationalized as quality, accessibility, and the cost of care. Methods and findings Literature to June 2021 was searched across five biomedical databases and filtered for eligibility. Publications that met the inclusion criteria were categorized by date, country, design, and results by three researchers independently. All studies were screened with the Risk of Bias in Non-randomised Studies—of Interventions (ROBIN-I) tool. The literature search produced 4,855 titles, and after applying criteria, 39 studies met inclusion (34 North America, 4 Europe, 1 Africa). Ten studies had a prospective design, and 29 were retrospective. Four studies were assessed as biased in results reporting. While most studies included a small number of PAs, five studies were national in origin and assessed the employment of a few hundred PAs and their care of thousands of patients. In 34 studies, the PA was employed as a substitute for traditional physician services, and in five studies, the PA was employed in a complementary role. The quality of care delivered by a PA was comparable to a physician’s care in 15 studies, and in 18 studies, the quality of care exceeded that of a physician. In total, 29 studies showed that both labor and resource costs were lower when the PA delivered the care than when the physician delivered the care. Conclusions Most of the studies were of good methodological quality, and the results point in the same direction; PAs delivered the same or better care outcomes as physicians with the same or less cost of care. Sometimes this efficiency was due to their reduced labor cost and sometimes because they were more effective as producers of care and activity.


Introduction
Healthcare systems across the globe face several challenges to meet patient demand and deliver high-quality healthcare. The challenges are primarily population growth, increasing chronically ill patients, rising patient expectations, and longevity [1]. At the same time, the supply of physicians is limited in many countries, leading to medical labor shortages [2].
The gap between physician capacity and patient demand is expanding and requires a change to the medical workforce. At least 18 countries have introduced the Physician Assistant/Associate (PA) as a solution [3]. These health professionals perform various medical and surgical services, and their numbers are growing across multiple settings.
In labor economics, if a PA replicates the activities of a physician, then that is a 'physician substitute' [4]. If, on the other hand, the PA improves the throughput in the medical process, then the PA is a 'complement' of physician services [5,6]. In most instances, the employment of the PA was the result of a medical shortage or a need to improve the quality of the medical service.
Because of their increasing utilization worldwide, understanding the economic value has become essential to their utilization. To date, no published systematic reviews have examined the cost-effectiveness of the PA. Therefore, the aim of this project was to review the effects of quality of care, accessibility of care, and costs of physician substitution by PAs in a variety of settings. This was operationalized as a research question: What is the cost-effectiveness of PAs compared to physician services?

Methods
A systematic review was undertaken using the reporting criteria developed at the University of York [7]. The protocol outlined the overview, objectives, aims, operational definitions, search strategy, inclusion/exclusion criteria, and quality appraisal methods.

Search strategy
The following international bibliographies were systematically searched: PubMed, Web of Science core collection (WoS), CINAHL (with full-text EBSCO), Embase-Ovid, and The Cochrane Library. A detailed search strategy was developed in consultation with two experts; a librarian experienced in systematic reviews and a health workforce researcher. The search strategy used PubMed as a format and then adapted it to the other database results. Searches were performed in 2021 and spanned all published studies through June 2021. Subsequently, the included articles and references were examined using a backward and forward snowball citation search method in Web of Science and Google Scholar to identify relevant other studies.

Inclusion and exclusion criteria
The literature search included all original empirical research studies on PAs with a comparative quantitative evaluation design written in English or Dutch. There were no date restrictions on publications. Both 'Physician Assistant' and 'Physician Associate' were included in the review, as they have a similar scope of practice. In addition, studies of 'Clinical Assistants' working in South Africa were included because their role is similar to, and modeled after, the PA [8,9].
Studies that encompassed nurse practitioners (NPs) and PAs but the provider type was missing were excluded. We omitted findings in which PAs were still in training, or the setting had an educational purpose. Articles were excluded when the outcome of care did not fit the protocol or where the care outcome of PAs was not compared to those of physicians.

Study selection
Citations from the systematic literature search were uploaded to the screening process to Rayyan QCRI, a systematic review computer-based application system [10]. Two of three reviewers screened all articles independently (GvdB, AvV, RSH) and were blinded to the others' findings. Abstracts were vetted using the inclusion/exclusion criteria, and ineligible reports were omitted. Those abstracts receiving conflicting votes were discussed, and after reading the text, consensus for inclusion or exclusion was reached. Articles were rejected when a PA and NP were included in the aggregate but not separated as two providers (and not compared one to the other).

Data collection, analysis, and synthesis
Two reviewers (GvdB, RSH), acting independently, extracted data from each article using a structured form and blinded to the other's findings. In addition, five corresponding authors of a candidate study were asked for clarifying information, such as the number of PAs in the project or how many clinics were involved.
Each article was assessed for quality using the Risk of Bias in Non-Randomised Studies-of Interventions (ROBIN-I) tool. The ROBIN-I instrument was developed for healthcare evaluation with potential biases in non-randomized studies that compare the effects of two or more interventions [11]. Assessing the risk of bias resulted in a summary score for every research domain ranging from 0 when there was no information; 1 for low risk of bias; 2 for moderate risk of bias; 3 for a significant risk of bias; and 4 for risk of bias was critical. When there was no information, the score was assessed as a serious risk of bias. These different scores per domain result in an overall risk of bias score from 1 to 4 (low bias to the critical risk of bias).
The first 19 data-extracted articles were reviewed by two reviewers independently, and a 97% agreement was reached for all criteria. Based on the high degree of agreement, the remaining articles were assessed by one reviewer (GvdB). The different scores per domain resulted in an overall risk of bias from 1 to 4 (low to critical risk of bias).
Extracted data were organized as: 1. General information (i.e., author, year of publication, country, setting).
3. Description of the intervention and whether the PA acted as a labor substitute or complement to a physician.
Papers that draw on the same study were extracted and analyzed as one study.
The following outcomes representing cost-effectiveness were assessed: Quality of care. The quality measurement of healthcare is based on the Donabedian model [12]. Metrics of quality of care are outcomes of care and the process of care. Evaluating the quality of care underpins the measurement for organizational improvement and is a primary focus of health services research [13].
Patient outcomes. these include morbidity, mortality, patient satisfaction, quality of life, health status, knowledge, and preference for a physician or PA.
Process of care outcomes. patient safety, quality of healthcare, adherence/compliance to guidelines or protocols, healthcare activities (examination, provision of advice, etc.), and referrals to other healthcare services.
Care provider (physician, PA) outcomes. includes workload (objective and subjective) and job satisfaction.
Accessibility of care. The focus on the accessibility of care is the employment effect of the PA on a patient entering the healthcare system. A component of access is the patient's waiting time to be seen for a medical or surgical condition.
Costs of care. Cost of care is the expenditures or utilization of resources in the delivery of healthcare services.

Results
In total, there were 4,855 titles of abstracts, papers, or reports identified by searching the bibliographies. After de-duplicating, 3,103 titles remained and were screened on title and abstract. The remaining records were assessed for the availability of a full report or article that was peerreviewed prior to publication. Many titles were poster or presentation abstracts without sufficient details on the methods and analysis and were excluded. After this screening, 54 articles remained, resulting in discussion and five instances of communicating with the author for more information. As a result of the final filtering process and discussion of each paper, a total of 42 articles emerged from the sorting process for final inclusion. The literature retrieval and study selection are shown in Fig 1.

Characteristics of included studies
The included articles (N = 42) spanned the years 1977-2021. The national origins of the articles were: North America (n = 34), Europe (n = 7), and Africa (n = 1). Over 500 PAs were involved or observed, and their numbers ranged from 1 to 443 (almost 50% of the studies reporting five or fewer). The quantity of PAs in the aggregate is unknown since the number was not consistently stated, and follow-up correspondence with authors did not often reveal more information. Study designs ranged from retrospective cohort studies (n = 29) to prospective studies (n = 10-including one time-motion study).

Description of the intervention
Throughout the studies, the effect of the introduction of PAs into the medical workforce was described. In most studies (n = 34), the economic labor effect was substitution-i.e., the PA produced a service that was (or had been) traditionally performed by a physician. The other five studies described a more complementary labor role where the PA enhanced the physician's role or improved the throughput of medical services [14][15][16][17][18].
Eight studies described the introduction of a PA for a single procedure [17][18][19][20][21][22][23][24]. In the remainder of the studies, the PA was assigned broad medical tasks. In five studies, the introduction of the PA was accompanied by organizational changes or the adaptation of various work processes [16,17,[25][26][27]. Changes included extra training or expanded time per patient, dedication to some procedure, more supervision by senior medical staff, or a combination of factors.
The characteristics of included studies are listed in Table 1.

Risk of bias in individual studies
Thirty-five of 39 studies in this review had a low risk of bias when assessed by the selection process, including missing data and results (See S1 and S2 Appendices for details). However, three studies [20,29,43] scored a serious risk of bias, and one study [17] scored a critical risk of bias in terms of confounding variables. The risk of bias scores is summarized and displayed in

PLOS ONE
The cost-effectiveness of physician assistants/associates Intervention: With the introduction of the PA in the ED, the PA saw 9,701 patients with the family practitioner (PA group).
Control: 10,776 patients who visit the ED are seen by a family practitioner (MD-group).
(Continued ) Intervention: PAs who worked as the PCP for residents in an NH.
Control: Physicians who worked as the PCP for residents in an NH.
(Continued ) Question: To assess the quality of patient care during the transition from resident-to PA-assisted trauma program (without residents) and simultaneous comparative support.
Intervention: 479 patients received care from PAs in a PA-assisted trauma program (without residents) and simultaneous comparative support.
Control: 293 patients received care from MD resident-assisted trauma program and simultaneous comparative support.

PLOS ONE
The cost-effectiveness of physician assistants/associates

PLOS ONE
The cost-effectiveness of physician assistants/associates

PLOS ONE
The cost-effectiveness of physician assistants/associates (Continued )

PLOS ONE
The cost-effectiveness of physician assistants/associates   (Continued ) quality of life [27]. Patient satisfaction of PAs did not significantly differ from the patient satisfaction of a physician in the three studies that reported this outcome. However, patients did not always distinguish that the PA was not a physician [16,25,51]. Process of care. In five studies, the process of care remained the same [19,25,27,31,35], and in four studies, the outcome improved with the addition of a PA [28,30,32,34]. Improvements were the use of thrombosis prophylaxis, beta-blockers, statins, or monitoring of blood pressure and blood glucose.
Provider outcomes. No studies reported the broader aspects of provider outcomes, such as workload or job satisfaction.
Accessibility of care. Four emergency department or acute care studies measured patient accessibility [15,16,28,30]. Three studies reported a decreased waiting time [15,16,28], and two studies showed a reduction in the proportion of patients leaving without being seen [15,30].
The cost of care, in monetary terms, measured in 11 studies, decreased with the introduction of a PA, or the results were equal to that of a physician alone (whether as a physician replacement or to improve the process of care) [14, 24-27, 29, 35, 38, 39, 51, 53. In one study, the cost of care by the PA was slightly greater than the physician's care [53]. In another case, the PA provided a financial benefit when the reimbursement was at least 80% of an MD's charge [47].
Two studies [20,22] researched the procedural times in cardiac angioplasty between cardiology fellows and a cardiology PA. The PA produced slightly faster procedure times with less fluoroscopic exposure time. Yang 2018 [40] Median hemoglobin A1c was comparable at diagnosis (6.6%, 6.7%, 6.7%, P > .05) and after 4 years (all 6.5%, P >.05). A1c levels at initiation of the first (7.5%-7.6%) and second (8.0%-8.2%) oral medications for patients of PA and NPs compared with that of physicians was also similar after adjusting for patient characteristics (all P > .05).
For the most part, the reviewed studies in Table 2 did not produce a significant 'differences of effect' analysis. We note that in two ambulatory studies, the employment of the PA was associated with a slightly longer patient LoV (by a few hours). However, the cost of patient care when delivered by a physician exceeded the cost of care provided by PA [25,29].
Three studies examined care outcomes by assessing cost-benefit and cost-utility-measuring the downstream cost-effectiveness of care or services [25,38,48]. In the Hooker 2002 study [48] and the Morgan 2019 study [38], the PAs did not negate their cost-benefit of less expensive labor by ordering more resources for an episode of care. In addition to the reduced labor cost, the medical resources used for an episode of care were less in the aggregate for the PA than the matched physician's resources for the same episode of care.
In five studies, the PA was employed not as a direct replacement for a physician but in response to increased demand for care [14][15][16][17][18]. Still, when added to the medical staff, the PA significantly improved the throughput of patient services (e.g., maxillofacial surgery, emergency department, or lung procurement for transplantation). In each instance, the inclusion of a PA resulted in time per patient saved. When a PA was introduced in a newly created fast track system in the emergency department, the 'through put' of patients improved, and patient waiting time decreased [16]. In these studies, no calculation was made of cost-effectiveness in terms of hospital, training, or healthcare costs at a national level. Nor were there any studies that researched the provider's workload or job satisfaction.

Discussion
This review of 39 studies involved synthesizing the evidence on the cost-effectiveness of PA employment. Thirty-two studies presented a retrospective data analysis. The majority of the research focused on a physician substitution effect (34 out of 39 studies). Five studies focused on the impact of PA employment along with their contribution to the efficient production of medical services [14][15][16][17][18]. While the retrospective studies were methodological sound, such ex post facto design is of lower grade than prospective ones. At the same time, higher levels of evidence, such as randomized controlled trials, are not often applicable as it is challenging to randomize healthcare workers since patients cannot be blinded to healthcare professionals.
Throughout the assessed reports, the question raised most often was whether the PA provided adequate care, cost-efficient care, or improved quality of care. In the aggregate, the costs of care were improved in 24 studies. In 16 cases, the quality of care was the same as that provided by a physician, and only in two studies did the visit time attributed to the PA lengthen [27,53]. In one study, the consult time of the PA slightly increased compared to the physician's consult time [25].
Rarely did these studies examine the broader organizational effect of whether the addition of a provider improves overall organizational efficiency. Drennan et al. point out that when the PA's service was incorporated in the cost-effectiveness analysis, this addition could have a broader impact on the cost of health services through referrals and prescriptions [25]. However, the authors concluded no significant differences in physician and PA rates of prescribing, ordering, referring, and consultation was found. As such, the costs were not assigned.
In terms of procedures, the outcome of circumcisions performed by a PA did not differ statistically from those of physicians. In contrast, the effects of performing surgical abortions, angioplasties, colonoscopies, and explanting lungs by PAs produced better outcomes compared with the physician's performance.
As a result of this systematic review, it is apparent that PAs are cost-effective in their delivery of patient care. Furthermore, their role as team members improved the quality of care through the input, throughput, or output. Although the labor cost of a PA versus a physician was implied in 15 studies, it was only categorically addressed in the Grzybicki [47], Hooker [49], and Timmermans study [27]. Aside from these examples, the implication is that physician employment cost and educational costs are higher than a PA.
The findings that emerge from this consolidated analysis are generalizable. They transcend five countries and represent the broad span of PA employment; acute care settings, medical and surgical wards, proceduralists, and facilitators of patient throughput. As a timeline, the published dates of the studies represent almost half a century of critical observation of PAs . The included studies offer a timeframe of cost-effectiveness of emerging roles of PAs and how their use expanded from their early introduction in small practices to contemporary medical centers in the 21 st century.
The first economic studies using a time-motion method to observe the interaction of PAs and physicians regarding patient care were in the USA and published at a time when the development of the profession was still in its infancy [6,14]. Early studies included some details of the PA, then known as a "new health professional," and drew on the limited literature known at the time [57].
In studies before the 1980s, the PA often worked in a protocol-driven context [58]. In observations after the 1980s, the PA profession was more established in healthcare and similar to today's professional profile, where the PA executes tasks independently. Their contemporary activity is viewed as an integrated member of a medical team [59].
From the 1990s onwards, the PA became more of a substitute for physician services in the role of a modern team player with a set of responsibilities [22,29,32,38,43,47,54]. By the new century, more countries had adopted the PA concept and drew on the American experience to develop their own professional PA profile [16,23,25,31,42].
Another observation of adding a PA was based on the decades of experience in the US and Canada for new PA adopters in Africa, Australia, and Europe. As the PA was considered in Europe in the new century, the implementers could draw on the experience, literature, demonstration studies, government reports, and observations of PAs at work to know how to best use their services and define their role [25]. By the second decade, the economics of their effectiveness had become more rigorous, as seen in the study of Timmerman and colleagues on costutility and Morgan's and colleague's studies of the cost-effectiveness of chronic disease management [38,56]. In essence, each team of researchers was able to sophistically account for the downstream effect of PA utility on 17 inpatient wards across the Netherlands and 170 VA medical centers with their associated 1,400 outpatient clinics.
When the various research questions posed in the included studies are analyzed, the PA's Scope of Practice (SoP) differed. Sometimes the PA's SoP was narrow; for example, independently performed surgical procedures as in circumcisions, lungs harvesting, surgical abortions, and cardiac catheterization. In other situations, they had broad medical tasks backfilling the physician's role on a ward or as an additional provider in an acute care setting with a commensurate SoP. In none of the articles did the researchers relate the SoP to the fourth goal in the "quadruple aim" of healthcare (i.e., taking care of health professionals) [60]. That raises the question of whether the performance of any procedure contributes to the experience of joy in their work as healthcare professionals [61]. However, the analysis of a half-century of PA job satisfaction literature suggests that almost all PAs find their role satisfying [62].
For the most part, the studies took place after the PA had been introduced into the organizational setting. In these situations, the outcomes before and after were compared. In five studies, the PA was added to a team [e.g., as part of a hospitalist service] or as a need to expand the medical staff [14][15][16][17][18]. Along with introducing a PA, organizational changes reflected on how services would be enhanced or improved. An example of organizational change is illustrated by Decloe et al. [42]. The PA was added to the infectious disease consulting service to mitigate the length of stay and patient morbidity and mortality in a Canadian hospital [42]. In another study, the medical residents that served as hospitalists were replaced with PA hospitalists in a small community hospital [43]. Both settings required significant organizational changes in staffing, hospital bylaws, on-boarding, and oversight of the PA.
We note that in the majority of studies in this review, the profiles of the PAs were missing. Most findings came up short on information as to the experience the PA brought to the setting. The exception was the de Lusignan study that noted the provider's gender and experience [44].
Supervision of the PA by the physician was considered a necessary activity, especially during the first decade or so of the introduction of the PA profession. When a supervising physician took time off from their patient schedule to supervise the PA's care or medical notes, the time was deducted from the PA's employment benefits [6]. In 11 studies, this variable was noted, but only one study calculated the economic effect [27]. Many studies indicated that when comparing medical or surgical residents and PAs, the supervision by an attending physician or senior consultant was equal. Two studies identified that the use of the PA saved time for the medical specialist without having operationalized it further [18,28].
Finally, we note that the effects of introducing a PA in several studies can be seen from the perspective of complex organizational change. The evaluation of a PA's introduction, often as a new health professional in the chain of care, is not the same as a treatment intervention. One of the first scholars of PA effectiveness noted: "As a theory, productivity is a simple concept: it measures changes in the total output that occurs when small changes are made in one factor of production, with all other factors and circumstances held constant. Because these conditions can be met in the real world only rarely, productivity numbers are almost always rough estimates. Certainly, that is the case concerning PAs." [6].

Limitations
One limitation of this analysis is that the settings and the outcome parameters differed across studies, and the characteristics of the PA were often missing. More granular PA and physician information is needed to understand what could be influencing or confounding variables that affect the actual outcome. Variables missing across almost all studies are the experience, educational level, number of involved PAs, and their age, gender, and background.
Another limitation was the need to separate the outcomes of the employment of the PA and NP. We omitted studies where the combined labor was not isolated. In five cases, we inquired whether the two providers could be separated for analysis. Understanding where the division of labor exists when three medical professionals work together is a health services research area that needs further exploration.
One strength of this systematic review was the reliance on peer-reviewed and published studies. As a result, various government-initiated PA demonstration projects promulgated as reports were excluded as not peer-reviewed [referred to as 'grey literature']. Another strength was the breadth of the search that provided clear insight into the PA profession's different effects and development. With the help of an experienced librarian, the research question was carefully operationalized. Combined with a reference check at the end of the process, the risk of missing relevant articles was significantly reduced.

Conclusion
The PA of the 21 st century is a semi-autonomous health professional who is a part of contemporary medical treatment teams. When peer-reviewed published studies spanning three continents were examined for quality of care, accessibility, and cost-effectiveness of employment, the PA was comparable to the physician in producing similar results in almost every case. Although some of the studies suggest that the addition of a PA resulted in a similar quality of care as physicians, in a few instances, their utilization enhanced the overall quality of care. In most instances, the introduction of a PA leads to the same or an improved quality of care, and their employment is cost-efficient when considering the labor and educational costs. These economic findings were observed in prospective and retrospective designs and various settings, whether primary care in outpatient offices or secondary [hospital-based] care. The results of the collective studies have produced a sizeable contextual understanding of efficient outcomes of care when the PA is a part of the medical team.