Research Special Paper
Impact of physician assistants on quality of care: rapid review
Nicola Cooper, Steven Agius, Kate Freeman, et al
BMJ 2025; 390 doi: https://doi.org/10.1136/bmj-2025-086358 (Published 03 July 2025)Cite this as: BMJ 2025;390:e086358
Abstract
Objective To determine the impact of physician assistants, compared with physicians, on quality of care in the context of an ongoing UK policy review.
Design Rapid systematic review.
Search strategy Keyword search of three databases; search and citation tracking of previous systematic reviews.
Eligibility criteria Empirical studies that quantitatively compared care delivered by physician assistants with care delivered by physicians, including residents, in economically developed countries, published between January 2005 and January 2025.
Main outcomes of interest Measures of outcomes of care, as defined by the Institute of Medicine’s definition of quality: safety, effectiveness, patient centredness, timeliness, efficiency, and equity.
Methods Eligible studies were categorised as primary care, secondary care, physician assistants versus residents in hospitals, diagnosis/performance, and cost effectiveness. Two reviewers independently extracted data on study design, samples, methods, and findings. Each study was assessed using a risk of bias tool. Owing to the heterogeneity of included studies, a narrative synthesis of the main findings was conducted. An assessment of confidence in the body of evidence for each outcome was based on the number and quality of relevant studies and the consistency of results between similar studies.
Results Of 3636 studies screened, 167 studies were eligible and 40 met the inclusion criteria. These consisted mainly of retrospective observational studies of weak quality. Most (32/40) were from the US, and no data from a post-covid-19 context were found. The greatest number of studies with the most consistent results were those that found that physician assistants practised safely and effectively when working under direct supervision and in post-diagnostic care. No difference was found in patient satisfaction between physician assistants and physicians. Although adding physician assistants to medical teams increases access to care, this may reflect the benefits of increased staffing rather than the unique contribution of the physician assistant role. Evidence on cost effectiveness is limited. Patients in the UK are more likely to see a physician assistant if they live in a socioeconomically deprived area.
Table 1
Summary of results—primary care
| Author, year | Country and setting | Study design and comparators | Data analysed | Measures | Results | Quality assessment |
|---|---|---|---|---|---|---|
| Marcum et al, 201632 | USA. Primary care | Retrospective cohort study. Adoption of new chronic disease medication. Physicians vNPs v PAs | All prescriptions of 5 new chronic disease medications dispensed in Pennsylvania between 1 Jan 2007 and 31 Dec 2011 | Three measures of prescriber adoption during 15 month post-FDA approval period: time to first prescription of new medication (speed of adoption); any prescription of new medication in final year of study period (extent of adoption); proportion of prescriptions of new medication in that class (extent of adoption) | >30 million prescriptions were analysed (data for each drug not presented here for brevity). Physicians adopted new drugs faster than NPs and PAs. Physicians prescribed larger share of new medications in final year of study. Physicians prescribed larger share of new medications in that class. P<0.001 for comparisons across providers for all variables | Weak. Non-randomised retrospective design. Confounders: possible differences in patient characteristics by provider type not accounted for |
| Jackson et al, 201833 | USA. Primary care (VHA) | Retrospective cohort study. Intermediate diabetes outcomes. Physicians vNPs v PAs | Review of 368 481 patient records meeting inclusion criteria: adults with diabetes treated with medication, with >1 primary care visit and >1 outpatient visit during study period with same provider | Patient characteristics. HbA1c control. Systolic blood pressure. LDL cholesterol concentrations. A priori thresholds were set for clinical significance of observed differences in clinical outcomes (0.3% for HbA1c, 3 mm Hg for SBP, and 0.13 mmol/L (5 mg/dL) for LDL cholesterol | Main care provider was provider most often visited at patient’s “home” clinic. These were physicians (n=3487), NPs (n=1445), and PAs (n=443) for 74.9% (n=276 009), 18.2% (n=67 120), and 6.9% (n=25 352) of patients, respectively. No statistically significant difference found in patient characteristics by provider type. No clinically significant differences found in HbA1c, systolic blood pressure, or LDL cholesterol concentrations by provider type | Weak. Non-randomised retrospective design. Confounders: team care and visits to other clinics not accounted for |
| Yang et al, 201834 | USA. Primary care | Retrospective cohort study. First 5 years of diabetes management. Physicians vNPs v PAs | Review of 19 238 records of patients with newly diagnosed diabetes who saw same provider type >75% of time, followed over 4 years | Patient characteristics. Prescriptions of oral antihyperglycaemic drug or insulin. Referral to diabetes clinic. HbA1c control | 78.2% (n=15 050) of patients managed by physicians, 14.7% (n=2821) by NPs, and 7.1% (n=1367) by PAs. Concordance with being seen by same provider was 92% for physicians, 82.3% for NPs, and 78.9% for PAs. No statistically significant difference in patient characteristics by provider type, apart from physicians statistically more likely to care for black/African American patients (P=0.02). Physicians referred 7.6% of patients to diabetes clinic, NPs 7.0%, and PAs 4.8% (P<0.001). No statistically significant differences in HbA1c at diagnosis, initiation of first and second oral drug, or after 4 years, after adjustment for patient characteristics. Statistically significant differences found in insulin prescriptions (physicians 14.6% of patients, NPs 14.1%, and PAs 12.5% (P=0.09) | Weak. Non-randomised retrospective design. Confounders: not clear whether or how often physician had input into NP and PA care. Race/ethnicity data not available for 10.1% of patients. Fewer insulin prescriptions and diabetes clinic referrals among PA cohort may suggest PAs saw less complex patients |
| Everett et al, 201935 | USA. Primary care (VHA) | Retrospective cohort study. Intermediate diabetes outcomes. Physician only vprimarily physician with NP vprimarily physician with PA vprimarily NP with physician vprimarily PA with physician vNP v PA | Review of 609 668 records of patients with diabetes assigned to different primary providers | Patient characteristics. HbA1c. Systolic BP. LDL cholesterol. A priori thresholds were set for clinical significance of observed differences in clinical outcomes (0.3% for HbA1c, 3 mm Hg for SBP, and 0.13 mmol/L (5 mg/dL) for LDL cholesterol | 66.9% (n=408 009) of patients managed by physicians only, 6.5% (n=39 861) by primarily physician with NP, 4.1% (n=24 692) by primarily physician with PA, 5.3% (n=32 472) by primarily NP with physician, 2.4% (n=14 342) by primarily PA with physician, 10.8% (n=66 042) by NP, and 4% (n=24 250) by PA. No statistically significant difference found in patient characteristics by provider type. No clinically significant differences found in HbA1c, systolic blood pressure, or LDL cholesterol concentrations by provider type | Weak. Non-randomised retrospective design. Confounders: team care not accounted for; patients with diabetes in VHA see their usual provider for around 75% of visits |
| Kurtzman and Barnow, 201736 | USA. Community health centres | Retrospective cohort study. Quality of care and practice patterns. Physicians vNPs v PAs | Review of 23 704 records of patient visits between 2006 and 2010 from National Ambulatory Medical Care Survey. Involving 1139 practitioners | Patient characteristics. Smoking cessation counselling. Depression treatment. Statin for hyperlipidaemia. Physical examination. Total No of health education/counselling services. Imaging. Total No of medications. Return visit at specified time. Referral to another physician | Physicians comprised 65% (n=742) of providers, NPs 26% (n=291), and PAs 9% (n=106). NPs were more likely to see female patients. NPs and PAs were more likely to see patients in rural areas (P<0.01 for both variables). Of 9 outcomes studied, only 2 had statistically significant differences. NPs were more likely to give recommended smoking cessation counselling (adjusted OR 1.62, 95% CI 1.17 to 2.26; P≤0.01). NPs and PAs were more likely to provide education/counselling services (adjusted incidence rate ratio 1.40, 95% CI 1.19 to 1.64; P≤0.01) | Weak. Non-randomised retrospective design. Confounders: potential differences in caseload (number of patients, throughput, and complexity) not accounted for |
| Drennan et al, 201537 | UK. Primary care | Retrospective cohort study. Outcomes and costs of consultations. PAs v GPs | Review of 2086 records of patients with same day appointments at 12 GP practices in England over 2 weeks in summer and 2 weeks in winter | Patient characteristics. Re-consultation within 14 days for same problem. Process measures: No of tests, referrals, prescriptions, general advice, medication advice. Length of consultation. Cost of consultation (calculated by salary and related costs of PA/GP plus consultation length). Patient satisfaction. Clinical review of records of re-consulting patients | PAs and GPs saw significantly different patients. PAs saw patients who were younger (mean age 34.45 (SD 23.2) v GPs 42.93 (24.87); P<0.001); had fewer chronic diseases (mean per patient 0.55 (SD 0.99) v GPs 0.87 (1.26); P<0.001); had fewer repeat prescriptions (mean 1.81 (SD 3.02) v GPs 2.60 (3.63); P<0.001); had fewer visits to practice in previous 3 months (mean 2.12 (SD 2.83) v GPs 2.70 (2.99); P<0.001); were more likely to live in deprived area (mean index of multiple deprivation 21.99 (SD 16.61) v GPs 15.81 (13.04); P<0.001); were more likely to present with minor problems or symptoms (62.9% of consultations v 50.2% for GPs; P<0.001), and were less likely to present with chronic problems (32.7% of consultations v 43.7% for GPs; P<0.001). PAs attended significantly more patients presenting for “minor problems or symptoms” and less often “chronic” problems than GPs. No significant difference in re-consultation rates between those who initially consulted PAs or GPs (including when adjusted for planned re-consultations) | Weak. Non-randomised retrospective design. Confounders: no data on time spent on supervision (direct or indirect) and signing prescriptions by GPs. Small number of PAs included in this study (n=7) |
| After adjustment for clustering at practice level, presenting problem, and patient characteristics described above, no significant difference was found between PAs and GPs in rate of diagnostic tests ordered, referrals to secondary care, or prescriptions issued. PAs were significantly more likely to document giving general advice (adjusted OR 3.30, 95% CI 1.69 to 6.45; P<0.001) and advice on medication management (adjusted OR 1.72, 1.08 to 2.73; P=0.02). Consultation times were significantly different. PA consultation time mean of 16.8 (SD 8.3) min v GPs 11.3 (7.6) min; P<0.001. GPs saw 3 patients for every 2 seen by PAs. Costs per consultation were lower for PAs (£28.14 v£34.36 for GPs). 93% of records of re-consulting patients were reviewed by GP reviewers for overall appropriateness, subjective information, objective information, documentation of assessment/problem, and plan (investigations, prescription). In all elements, PAs performed better than GPs, with overall appropriateness of 81.6% v 50.8% of GPs. GP reviewers incorrectly judged 58.3% of PA consultations to be those of GPs and 23% of GP consultations to be those of PAs |
CI=confidence interval; FDA=Food and Drug Administration; GP=general practitioner; LDL=low density lipoprotein; NP=nurse practitioner; OR=odds ratio; PA=physician assistant; SBP=systolic blood pressure; SD=standard deviation; VHA=Veterans’ Health Administration.
Table 2
Summary of results—secondary care
| Author, year | Country and setting | Study design and comparators | Data analysed | Measures | Results | Quality assessment |
|---|---|---|---|---|---|---|
| Kurtzman et al, 202338 | USA. Emergency department | Retrospective cohort study. Practice patterns of teams in EDs. Physician v NP vPA led teams | Review of 95 718 records—randomly sampled patient visits within 12 year period (2009-20) from National Hospital Ambulatory Medical Care Survey | Patient characteristics. Patient clinical information. Day and time of visit. Patient disposition (eg, transferred, admitted, died in ED). Patient flow indicators (ie, arrival time, time seen, time discharged). Process measures: diagnostic tests, procedures, prescriptions | Physicians saw patients who were statistically significantly older (47.1% aged >45 years v 33.2% for NPs v 35.2% for PAs; P≤0.01); more severely ill on basis of triage level (8.4% v 2.0% for NPs v 2.4% for PAs; P≤0.01); more often seen between midnight and 8 am (16.4% v8.4% for NPs v 7.2% for PAs; P≤0.01) and transported by ambulance (12.5% v 4.5% for NPs v 4.0% for PAs; P≤0.01). PAs and NPs were statistically significantly more likely to see patients presenting with injuries and poisoning (30.5% for physicians v 38.9% for NPs v 41.4% for PAs; P≤0.01) | Weak. Non-randomised (to intervention) retrospective design. Confounders: team care not accounted for |
| Multivariate regression analysis was used to adjust for patient and clinical characteristics to examine associations between provider type and outcomes of interest. Being seen by PA team significantly lowered adjusted incident rate ratio of diagnostic services (0.8, 95% CI 0.7 to 0.8) and procedures (0.9, 0.8 to 1.0). PA teams were associated with significant decrease in length of a visit by 26.3 (95% CI −36.7 to−15.9) min and reduction in odds of hospital admission (adjusted OR 0.3, 95% CI 0.2 to 0.4) | ||||||
| Moore et al, 202139 | USA. Emergency department | Retrospective cohort study. ED. Physicians vPAs | Review of 25 883 records of patients who attended between Apr 2016 and Dec 2018 | Patient characteristics. Patient acuity. Mean LOS. Door to provider time. Re-attendance within 72 h. No of patients seen per hour. No of CT scans requested. Patient disposition. Patient satisfaction | 58.7% (n=15 205) of patient encounters were managed by PAs and 41.3% (n=10 678) by physicians. Statistically significant differences were found in patient characteristics, with patients cared for by physicians being older (36.4 (SD 23.5) vPAs 34.4 (22.7) years; P<0.001) and physicians caring for greater proportion of patients aged >65 (14.2% v 11.8% for PAs; P<0.05). No clinically significant differences were found in patient acuity by provider type. Patients cared for by PAs had longer mean LOS (126 (SD 96) v 120 (96) min for physicians; P<0.001); and longer door-to-clinician times (16.2 (SD 66) v 11.4 (108) min for physicians; P<0.001). No statistically significant differences found in re-attendance rates, No of patients seen per hour, and CT scans requested. Patient satisfaction was statistically significantly higher for PAs (mean score 9.3 (SD 0.2) v 9 (0.2) for physicians; P<0.001) | Weak. Non-randomised retrospective design. Confounders: team care not accounted for; no adjustments made for differences in patient characteristics |
| Pavlik et al, 201740 | USA. Emergency department (paediatric patients) | Retrospective cohort study. Physicians v PAs | Review of 10 369 records of children who attended general community ED in 24 month period | Patient characteristics. Unplanned re-attendances within 72 h | Physicians were statistically significantly more likely to see youngest patients, demonstrated graphically by showing linear correlation between younger age and physician care (85% of 8 week olds falling to 42% of 6 year olds seen by physicians; P<0.001). Physicians saw most patients, but proportion of patients seen by acuity was significantly different, with physicians seeing greater proportion of patients with higher acuity scores and PAs seeing greater proportion of patients with lower acuity scores (P<0.001). No statistically significant difference was found in re-attendance rates by provider. Patients for whom PAs sought input from attending physicians had highest re-attendance rates | Weak. Non-randomised retrospective design; no adjustments made for differences in patient characteristics. |
| Halter et al, 202042 | UK. Emergency department | Mixed methods study: retrospective records review; semi-structured interviews; observation of staff. Processes and outcomes of consultations. PAs v FY2 doctors | Review of 8816 records of patients who attended in 16 week period. 3 EDs. 6 PAs and 40 FY2 doctors. 14 clinicians/managers and 6 patients/relatives interviewed. 5 PAs observed | Patient characteristics. Patient acuity. Area seen. Primary outcome: unplanned re-attendance within 7 days. Secondary outcomes: prescriptions given, admission to hospital, radiography requests, discharge summary completed. 40 records selected for detailed review for clinical adequacy of documented care | FY2 doctors were statistically significantly more likely to see older patients (P=0.002, all age groups) and patients triaged as urgent/very urgent (53.1% for PAs v66.8% for FY2s; P<0.001) but less likely to see patients triaged as immediate (0.6% for PAs v 0.1% for FY2s; P<0.001). FY2 doctors were also more likely to see patients in majors (68.8% for PAs v 88.4% for FY2s; P<0.001) and less likely to see patients in minors (20.1% for PAs v 6.8% for FY2s; P<0.001). PAs did not see any patients overnight but FY2 doctors did | Weak. Non-randomised retrospective design. Confounders: no adjustments made for differences in area of ED seen, day, and time of day |
| After adjustment for patient age, sex, and triage score, no statistically significant differences were found in re-attendance rates. No statistically significant difference was found in prescriptions, admission rates, or discharge summary completed. Patients seen by PA were statistically significantly more likely to have radiography (adjusted OR 2.7, 95% CI 1.72 to 4.24; P<0.001) and had mean 35 min shorter LOS (no statistics provided) after adjustment for age, sex, acuity, whether admitted, radiograph taken, and hospital site. Clinical reviewers found 3/40 patient charts clinically inadequate (1 PA and 2 FY2). Patients were positive about care they had received from PA but had poor understanding of role | ||||||
| King et al, 202443 | UK. Emergency department | Retrospective cohort study. Outcomes of consultations. PAs v FY1 doctors | Review of 7405 records of adult patients who attended single ED between Aug 2018 and Jan 2020. 11 PAs and 7 FY1 doctors | Patient characteristics. Area seen. Day and time of day. Patient acuity. Primary outcome: wait time to consultation. Secondary outcomes: LOS in department, left without being seen, unplanned re-attendance within 72 h | FY1 doctors were statistically significantly more likely to see younger patients (all age groups P<0.001) and during normal working hours (81.8% patients seen 8 am to 4 pm for FY1s v 58.6% for PAs; P<0.001). FY1s were more likely to see patients in urgent treatment centre (55.3% of patients v 11.6% for PAs; P<0.001) and less likely to see patients in majors (42.8% of patients v 86.2% for PAs; P<0.001) | Weak. Non-randomised retrospective design. Confounders: team care not accounted for |
| After adjustment for patient age, time of day, area of ED seen, and patient disposal, no statistically significant difference was found in wait time to consultation. No statistically significant differences were found in left without being seen rates and unplanned re-attendances. Patients seen by PAs had statistically significantly longer LOS (adjusted mean 258.25 (95% CI 251.59 to 264.90) min v 198 (190.36 to 205.63) min for FY1s; P<0.001). Patients who left without being seen and those who were admitted were excluded from LOS calculations |
CI=confidence interval; CT=computed tomography; ED=emergency department; FY=foundation year; LOS=length of stay; NP=nurse practitioner; OR=odds ratio; PA=physician assistant; SD=standard deviation.
Table 3
Summary of results—residents versus physician assistants (PAs) in hospitals
| Author, year | Country and setting | Study design and comparators | Data analysed | Measures | Results | Quality assessment |
|---|---|---|---|---|---|---|
| Divi et al, 202144 | USA. Orthopaedic institute | Retrospective cohort study. Surgical outcomes. PAs vPGY2-5 (residents) and PGY6-7 (fellows) as assistants | Review of 171 records of patients undergoing lumbar decompression surgery | Patient characteristics. Total operative time. Readmission rates at 30 and 90 days. Need for revision surgery at 1 year. Postoperative PROMs | No statistically significant differences in patient characteristics between groups. Mean operative time was similar in fellow/resident group and PA group (179 v 188.5 min; P=0.58). No statistically significant difference in readmission rates. No statistically significant difference in revision surgery at 1 year. No statistical difference in postoperative PROMS | Weak. Non-randomised retrospective design. Confounders: different assistants assisted according to their level of ability |
| Hazzard et al, 202245 | USA. Department of orthopaedic surgery | Retrospective cohort study. Surgical outcomes. One experienced PA v rotating fellows as first assistants | Review of 264 records of patients undergoing anterior cruciate ligament reconstruction | Patient characteristics. Skin-to-skin time. Tourniquet time. Month of surgery. PROMs before and after surgery (up to 2 years) | No statistically significant differences in patient demographics between groups. Longer surgical times were seen in all 4 quarters of year for fellows, but improved each quarter until no statistically significant difference was found between rotating fellows and PA. Skin-to-skin surgical time during quarter 1 (15.9 min difference; P=0.02), quarter 2 (15.8 min difference; P=0.001), quarter 3 (12.1 min difference; P=0.001), and quarter 4 (4.4 min difference; P=0.20). Same pattern was seen for tourniquet time. No statistically significant differences in preoperative and postoperative PROMS between groups | Weak. Non-randomised retrospective design. Confounders: fellow instruction during surgery was not measured. |
| Malloy et al, 202146 | USA. Department of plastic surgery | Retrospective cohort study. Surgical outcomes. One PA with ≥2 years’ training v15 PGY3+ residents rotating every 2 months as first assistants | Review of 49 records of paediatric patients undergoing reduction mammaplasty | Patient characteristics. Financial data. Duration of surgery (“indirect cost”) | No statistically significant differences in patient characteristics were found between groups. Hospital costs were lower in PA group: median $28 997 (range $24 767-$39 775) for PA v $32 747 ($25 121-$43 753) for residents (P<0.01. Mean duration of surgery was shorter in PA group: mean 158 (SD 20) min for PA v mean 192 (28) min for residents (P<0.001) | Weak. Non-randomised retrospective design. Confounders: patients elected whether to stay overnight or not, affecting costs. Resident instruction during surgery was not measured |
| Divi et al, 202147 | USA. Department of orthopaedic surgery | Retrospective cohort study. Surgical outcomes. PA v PGY5-6 resident/fellow as first assistants | Review of 350 records of patients undergoing lumbar fusion surgery | Patient characteristics. Type of surgery. Total surgery time. Length of stay. Wound infection. Need for revision surgery at 1 year. PROMs | No statistically significant differences in patient characteristics, apart from the Charlson Comorbidity Index which was higher in the resident/fellow group: mean 3.40 (95% CI 3.07 to 3.74) v 2.69 (2.39 to 3.00) for PAs (P=0.002). No statistically significant difference in type of surgery, surgery time, length of stay, wound infection rates, need for revision surgery, and preoperative and postoperative PROMs | Weak. Non-randomised retrospective design. Confounders: patient groups differed in comorbidity score |
| Costa et al, 201348 | USA. Department of thoracic surgery | Retrospective cohort study. Surgical outcomes. PA as senior surgeon vfellow as senior surgeon, both working with residents | Review of 287 lung procurements for transplant performed between 2008 and 2012 | Surgical injuries to donor lung | Significant difference was seen in number of injuries during procurements, with PA having 1/197 (0.5%) and fellows having 22/90 (24%) (P<0.01). No statistically significant difference seen in 30 day and 1 year graft survival rates. Rates for pulmonary graft dysfunction grade 2-3 were significantly lower in PA cohort: 19/197 (9.6%) v 29/90 (32.2%) for fellows (P<0.01). Rates for pulmonary graft dysfunction grade 0-1 were significantly lower in fellow cohort: 61/90 (67.8%) v178/197 (90.4%) for PAs (P<0.01) | Weak. Non-randomised retrospective design. Confounders: unknown whether any differences in technical difficulty of procurement between two groups, as not measured |
| Quanbeck et al, 202549 | USA. Department of orthopaedic surgery | Retrospective cohort study. Surgical outcomes. Surgeon alone vPA, resident, or fellow as first assistant | Review of 888 cases of closed reduction and percutaneous pinning of paediatric supracondylar humerus fractures | Patient characteristics. Operating time. Complication rate | Other than fewer type II fractures for fellows, no significant difference were found in four groups relative to patient age or percentage of fracture type. Operating time was longer for more complex fractures for all groups. For all fracture types, surgeon operating alone took mean of 36 (SD 15.1) min, with PA 34.7 (13.5) min, with fellow 37.1 (11.6) min, and with resident 44.3 (23.6) min. Only operating with resident was statistically significant (P<0.001) with an 8.3 min increase compared with surgeon alone. No statistically significant difference was found in number of K wires placed or the length of follow-up on basis of assistant category | Weak. Retrospective design. Confounders: 134 patients lost to follow-up. Resident direct involvement and instruction during surgery not measured. |
| Lui et al, 202350 | USA. Department of orthopaedic surgery | Retrospective cohort study. Surgical outcomes. PA vresident v fellow v resident + fellow as first assistants | Review of 274 records of patients with cubital tunnel syndrome who had primary cubital tunnel surgery | Patient characteristics. Operating time. Complication rates. Length of stay | No statistically significant differences in patient characteristics between groups. Statistically significant difference was found in type of cubital tunnel surgery performed: no difference in “in situ” release; PAs more likely to perform subcutaneous transposition (39.5% v residents 13.2% v fellows 19.7% vresident/fellow 15.4%; P=0.008); fellows more likely to perform submuscular transposition (21.2% v PAs 5.3% v residents 15.4% vresident/fellow 0%; P=0.04). No association between surgical assistant and complications, reoperations, or length of stay | Weak. Retrospective design. Confounders: different types of surgery performed; subcutaneous transposition is technically easier but no technique is considered superior |
| Singh et al, 201151 | USA. General medicine inpatient service | Retrospective cohort study. Hospital outcomes. PA-hospitalist teams vresidents-attending teams | Review of 9681 general medical hospital records of patients admitted on weekdays between Jan 2005 and Dec 2006 | Patient characteristics. LOS. Costs. Unplanned readmission within 7, 14, and 30 days. Inpatient mortality | No statistically significant differences in patient characteristics or admission diagnosis between groups. Overall LOS was higher for PA-hospitalist teams by 6.73% (95% CI 1.99% to 11.70%; P<0.005). This persisted when winsorised data were used, and admissions that involved ICU stay and deaths were excluded. Analysis of subset of admissions during times when both types of teams could receive patients (11 am to 4 pm) showed no significant difference in LOS: 2.97% (95% CI −4.47% to 10.98%) higher (P=0.44). On restricting multivariable analyses to subset of hospitalists who staffed both types of teams, the increase in LOS associated with PA-hospitalist care was no longer significant: 5.44% (95% CI −0.65% to 11.91%) higher (P=0.08). No statistically significant differences in hospital charges, readmission at 7 and 30 days, and inpatient mortality after adjustment for patient characteristics, admission source, ward, time, day of week, and comorbidity measures | Weak. Non-randomised retrospective design. Confounders: severity of illness not accounted for |
| Glotzbecker et al, 201352 | USA. Chemotherapy service in academic medical centre | Retrospective cohort study. Patient outcomes. PAs vhouse officers, both working with attending physicians | Review of 95 records of patients admitted for chemotherapy for acute myeloid leukaemia between 2008 and 2012 | Patient characteristics. Primary outcomes: LOS, inpatient mortality. Secondary outcomes: readmission within 14 days, No of consults requested, ICU transfer | No statistically significant differences in patient characteristics between groups. Mean LOS was statistically significantly shorter in PA group: 30.9 days v 36.8 days for house officers (P=0.03). No statistically significant difference in inpatient mortality between groups. Readmission rates were statistically significantly lower in PA group: 0% v 10.6% for house officers (P=0.03). No of consults requested was also lower: mean per patient 1.47 v 2.11 for house officers (P=0.03). No statistically significant differences in ICU transfers | Weak. Non-randomised retrospective design. Confounders: practice patterns of attending physicians not accounted for |
| Roy et al, 200853 | USA. General medicine inpatient service | Retrospective cohort study. Hospital outcomes. PAs working with hospitalists vresidents working with attending physicians | Review of 5194 records of patients admitted between July 2005 and June 2006 | Patient characteristics. LOS. Hospital costs. Unplanned re-admission at 72 hours. 14 days, and 30 days. ICU admission. Inpatient mortality. Patient satisfaction | Patients in PA-hospitalist group were statistically significantly younger with lower Charlson Comorbidity Index scores (P=0.04 for all age groups; P=0.02 for all CCI scores). Patients in PA-hospitalist group were more likely to be admitted at night (43.8% v30.3% for residents + attendings; P<0.001). After adjustment for age, race, CCI, time of admission, insurer, and case mix index, no statistically significant difference was found in LOS, hospital costs, readmission rates, or in-patient mortality. No statistically significant difference in patient satisfaction scores was found between groups | Weak. Non-randomised retrospective design. Confounders: PA-hospitalist group were geographically localised; practice patterns of hospitalists/attending physicians not accounted for |
| Kawar and DiGiovine, 201154 | USA. Medical intensive care unit | Retrospective cohort study. Hospital outcomes. PAs in one ICU vresidents in another ICU, both groups working with critical care fellow/attending physician | Review of 5346 records of patients admitted to one of two medical ICUs in single hospital between Jan 2004 and Jan 2007 | Patient characteristics. ICU LOS. Hospital LOS. ICU mortality. Hospital mortality. Readmission rates | Resident group had higher rate of renal insufficiency (25% v 22%; P=0.05) and PA group had higher rate of stroke (5.6% v 4%; P=0.02); other patient characteristics (eg, age) were similar. After exclusion of patients who died in ICU, ICU LOS was longer for patients admitted to PA group (median 2.58 (95% CI 1.55 to 4.86) days) than for patients admitted to resident group (2.33 (1.39 to 4.16) days). This difference persisted after adjustment for severity of illness. No statistically significant difference was found in hospital LOS, ICU mortality, hospital mortality, or readmission rates. A matched analysis was done (identifying 1249 pairs of patients to create two matched groups). Using this well matched subgroup of patients, no significant differences were found in any outcomes | Weak. Non-randomised retrospective design. Confounders: PA ICU was covered by critical care fellows overnight |
| Dhuper and Choksi, 200955 | USA. General medicine inpatient service | Before and after study. Hospital outcomes. PAs vresidents, both working with attending physicians | Review of 5508 records of patients admitted to PA-hospitalist service (1998-2000) compared with 5458 records of patients admitted during residency-attending service (1996-98) | Mortality. Adverse events. Readmissions within 30 days. Patient satisfaction. Medical record review of all deaths and readmissions to identify any deficiencies in care | No data on patient characteristics between groups. Statistically significant reduction in all cause mortality found: 148/5508 (2.7%) deaths in PA-hospitalist model v235/5458 (4.3%) deaths in resident-attending model (P<0.001). Statistically significant reduction in case mix index adjusted mortality: 0.019 for PA-hospitalist model v 0.029 for resident-attending model (P<0.001). No statistically significant differences found in adverse events, readmissions, and patient satisfaction. Medical record reviews of readmissions found no statistically significant difference in incidences of inadequate care between groups | Weak. Non-randomised retrospective design. Confounders: changes in hospital or attending practices/new drugs/technology between 1996 and 1998 not accounted for (in 1988 institution went through major restructuring). PA-hospitalist group had 24/7 onsite supervision from hospitalist and intensivist |
| Bos et al, 201856 | Netherlands. Inpatient medical and surgical wards | Multicentre non-randomised, matched controlled study. Prescribing quality. Wards run by PAs >51% of time during 8 am to 6 pm weekdays vwards run by residents, both groups supervised by physicians | Review of 2307 records of adult inpatients on participating wards. | Patient characteristics. Hospital type. Ward type. Admission type. Primary diagnosis. Charlson Comorbidity Index score. 17 quality indicators to measure adherence to prescribing guidelines | No statistical analysis done on patient characteristics, hospital/ward/admission type, admission diagnosis, and comorbidity score to look for differences. 2/17 quality indicators showed statistically significantly less non-adherence for PA model. These were gastric protection in NSAID use in combination with corticosteroids (OR 0.42, 95% CI 0.19 to 0.90) and in use of NSAIDs in patients >70 years. No statistically significant differences were found in any other quality indicators for non-adherence to guidelines on medication prescribing. Adherence to recommendations varied across indicators but tended to be low overall | Weak. Non-randomised retrospective design. Confounders: wards not run by PAs 100% of time, team based care, practice patterns of supervisor not accounted for. PAs were ward based whereas residents had other duties |
| Timmermans et al, 201757 | Netherlands. Inpatient medical and surgical wards | Multicentre non-randomised, matched controlled study. Hospital outcomes. Wards run by PAs >51% of time during 8 am to 6 pm weekdays vwards run by residents, both groups supervised by physicians | Review of 2307 records of adult inpatients on participating wards. | Patient characteristics. Primary outcome: LOS. Secondary outcomes: quality and safety of care indicators: (eg, in-hospital mortality, unplanned admission to ICU, CPR, pressure ulcer development, hospital acquired infections, unplanned readmission). Patient experience | Uses same dataset as above study. Baseline characteristics of patients differed in specialty, hospital type, major diagnostic group, type of admission (elective v urgent), discharge destination, and ward workload (P<0.001). After adjustment for confounders (eg, patient characteristics), no statistically significant differences were found in LOS or any quality and safety of care indicators. Patient experiences of care (communication, continuity, cooperation, and medical care) were all rated as significantly higher in PA model (overall evaluation score 8.4 (SD 1.3) for PA group v 8.0 (1.5) for MD group; P<0.05) | Weak. Non-randomised retrospective design. Confounders: wards not run by PAs 100% of time, team based care and practice patterns of supervisor not accounted for. PAs were ward based whereas residents had other duties |
CI=confidence interval; CPR=cardiopulmonary resuscitation; FY=foundation year; ICU=intensive care unit; LOS=length of stay; NSAID=non-steroidal anti-inflammatory drug; OR=odds ratio; PA=physician assistant; PGY=postgraduate year; PROM=patient reported outcome measure; SD=standard deviation.
Table 4
Summary of results—diagnosis/performance
| Author, year | Country and setting | Study design and comparators | Data analysed | Measures | Results | Quality assessment |
|---|---|---|---|---|---|---|
| Jiao et al, 201858 | USA. Ambulatory care | Retrospective cohort study. Prescribing practices. Physicians v NPs vPAs | Review of 701 499 records from two large databases of patient visits between 2006 and 2012 (96.8% of visits to physicians, 1.6% PAs, 1.6% NPs) | Patient characteristics. Patient payment source. 13 validated outpatient quality indicators for prescribing | Patient characteristics statistically significantly differed in all variables by clinician group (P<0.001). For example, physicians were more likely to see older, privately paying patients and PAs provided more care in outpatients and in emergency department. Overall mean performance across all indicators was 58.7%. After adjustment for potentially confounding patient, provider, and visit characteristics, no statistically significant differences were found in quality of prescribing practices between physicians and non-physicians (NPs and PAs) for 10/13 quality standards evaluated. Both NPs and PAs met quality standard of antithrombotic therapy for atrial fibrillation more often than physicians (OR 1.76, 95% CI 1.16 to 2.67) but were less likely to meet quality standards for treating depression (OR 0.75, 0.61 to 0.93) and antibiotic use in otitis media (0.41, 0.24 to 0.70) | Weak. Non-randomised retrospective design. Confounders: patient characteristics significantly differed by clinician type, diagnosis assumed to be correct |
| Ellenbogen and Segal, 202059 | USA. Primary care, urgent care, hospital discharges | Retrospective cohort study. Opioid prescribing. Physicians v NPs vPAs | All generalist physicians, NPs, and PAs who provided more than 10 Medicare Part D prescription claims between 2013 and 2016 (n=36 999) | Practice setting. Provider gender, years in practice. Practice setting and median income of ZIP code. Total prescription claims. Opioid claims as proportion of all prescription claims | Distributions of total opioid claims for all groups were extremely right skewed but clustered around zero (ie, relatively small No of practitioners were responsible for large proportion of prescriptions). Adjusted total number of opioid claims across study period was 660 (95% CI 660 to 661) for physicians, 755 (753 to 757) for NPs, and 812 (811 to 814) for PAs. NPs and PAs made up disproportionately high number of prescribers with highest 5% of opioid prescription proportions. PAs made up 43% of this group and 12% of entire study sample. NPs made up 32% of this group and 22% of study sample. Physicians made up 24% of this group and 66% of study sample. Significantly more data were withheld from public release for low prescription counts for NPs and PAs than for physicians. Percentage of physicians in year with withheld data ranged from 13.1% in 2015 to 14.1% in 2013. For NPs, it ranged from 20.0% in 2016 to 23.7% in 2014. For PAs, it ranged from 23.4% in 2015 to 25.2% in 2014 | Weak. Non-randomised retrospective design. Confounders: differences in percentages of withheld data between groups, differences in practice patterns, some states place restrictions on opioid prescribing by non-physicians |
| Lozada et al, 202060 | USA. Primary care | Retrospective, cohort study. Opioid prescribing. Physicians v NPs vPAs | Review of 222 689 Medicare Part D prescriptions (20% sample of all 2015 prescriptions in primary care) | Patient characteristics. State. Among providers who issued ≥50 prescriptions, identification of potential opioid overprescribing, as defined by one of: opioid prescribed to >50% of patients (high frequency); prescribed ≥100 morphine milligram equivalents (MME)/day to >10% of patients (high dose); prescribed opioid for >90 days to >20% of patients (long term) | 3.8% of physicians, 8.0% of NPs, and 9.8% of PAs overprescribed opioids. Most NPs and PAs prescribed opioids in pattern similar to physicians, but NPs and PAs had more outliers who prescribed high frequency, high dose opioids, after adjustment for patient comorbidity in multivariable analyses. Odds ratios were as follows: 2.96 (95% CI 2.78 to 3.15) for NPs, 5.73 (5.35 to 6.13) for PAs for high frequency opioids compared with physicians; 1.66 (1.52 to 1.80) for NPs, 2.16 (1.97 to 2.38) for PAs for high dose opioids compared with physicians. NPs and PAs were less likely than physicians to prescribe long term opioids: OR 0.57 (95% CI 0.53 to 0.61) for NPs, 0.71 (0.65 to 0.77) for PAs. NPs and PAs who practised in states with independent prescription authority were around 28 times more likely to overprescribe opioids than those in states that restricted authority | Weak. Non-randomised retrospective design. Confounders: palliative care patients not analysed separately, only 20% of claims analysed |
| Satyaprakash et al, 200761 | USA. Ambulatory care | Retrospective cohort study. Prescribing practices. Dermatologists vdermatology PA vprimary care physician v primary care PAs v other physician | Review of 301 million records of outpatient visits from National Ambulatory Medical Care Survey between 1995 and 2004 | Patient age, sex, and race/ethnicity. Prescribing rates of clotrimazole-betamethasone in different clinician groups, PAs with and without direct supervision by physician. | In multivariate logistic regression analyses, rates of prescribing clotrimazole-betamethasone, by clinician: PAs in primary care 16.9%; PAs in primary care with direct supervision 8.3%; primary care physician 4.9%; dermatology PAs 3.8%; dermatology PAs with direct supervision 1.1%; dermatologists 0.2%; other physicians 1.7%. Clotrimazole-betamethasone was more likely to be prescribed at visits to PAs (regardless of specialty) when PA was sole provider of dermatological care v when the physician assistant was under direct supervision by physician (OR 4.3 (95% CI 0.7 to 25.6) v 1.8 (0.4 to 8.0) | Moderate. Non-randomised retrospective design. Unclear whether patient characteristics were statistically significantly different between groups (but vast database) |
| Fejleh et al 202063 | USA. Veterans’ Affairs medical centre | Retrospective cohort study. Screening colonoscopy outcomes. PAs vfellows vgastroenterologists | Retrospective review of 597 records of patients undergoing average risk screening colonoscopies between July 2015 and June 2016 | Patient characteristics. Endoscopist experience level. Colonoscopy data (eg, caecal intubation, time to intubation). Adenoma detection rates | Study involved 5 PAs and 7 gastroenterologists (No of fellows unknown). No statistically significant differences in patient age, sex, and race/ethnicity between groups. PAs performed better than gastroenterology fellows with regard to mean intubation time (7.8 v 13.2 min; P<0.001) and had shorter mean withdrawal time (9.6 v 11.5 min; no P value given). No significant difference was found between mean intubation time of PAs and all attending gastroenterologists (7.8 v8.8 min; P=0.25). PAs with ≥15 years’ experience had shorter mean intubation times than 2 attending gastroenterologists with similar experience (15.6 v 7.5 min; P=0.002). PAs had higher caecal intubation rates than attending gastroenterologists (98.8% v 94.8%; P=0.04), but no difference was found when PAs were compared with fellows. PAs achieved average adenoma detection rate of 46.7%, which was comparable to both gastroenterologists (adenoma detection rate of 43.5%; P=0.59) and fellows (44.2%; P=0.89) | Weak. Non-randomised retrospective design. Confounders: no information on how patients were prospectively allocated to different clinician groups, poor bowel preparation excluded from analysis, gastroenterology fellows could be assisted by gastroenterologist, only 200 cases with no biopsy taken were analysed for withdrawal time |
| Anderson et al, 201864 | USA. Dermatology | Retrospective cohort study. Dermatology PAs vdermatologists | Review of 20 270 records of patients attending for skin screening over 5 years between 2011 and 2015 | Patient characteristics. Number needed to biopsy to diagnose skin cancer | No difference in patient characteristics, except patients with history of melanoma were more likely to see dermatologist, whereas those with history of any type of skin cancer were more likely to see PA (P<0.001 for both). PAs did more biopsies overall (22.9% of visits v20.8% for dermatologists; P<0.001) and of pigmented lesions (12.9% of visits v 11.1% for dermatologists; P<0.001). To diagnose 1 case of skin cancer, NNB was 3.9 for PAs and 3.3 for dermatologists (P<0.001). Per diagnosed melanoma, NNB was 39.4 for PAs and 25.4 for dermatologists (P=0.007). Patients screened by PA were significantly less likely than those screened by dermatologist to receive diagnosis of melanoma in situ (0.2% v 0.4% of visits; P=0.04), but differences were not significant for invasive melanoma (0.2% v 0.2% of visits; P>0.99) or non-melanoma skin cancer (6.1% v 6.1% of visit; P=0.98) | Weak. Non-randomised retrospective design. No strategy used to deal with statistically significant differences in patient characteristics |
| Brock et al, 201665 | USA. All settings | Retrospective cohort study. Physicians v NPs vPAs | Review of 178 035 malpractice claims in 10 year period from National Practitioner Data Bank | No of claims by clinician group. Type of claim by clinician group | Data included 104 482 unique providers: 94.8% (n=99 070) physicians, 2.9% (n=3064) PAs, and 2.2% (n=2256) NPs. Physicians had significantly more malpractice reports than adverse event reports (63.0% v 37.0%; P<0.001), but this relation was reversed for PAs, who had significantly fewer total malpractice reports than adverse event reports (28.1% v 71.9%; P<0.001). Across 10 year period, highest rate of malpractice reports for physicians was in 2005 (19.0 per 1000) and lowest in 2014 (11.2 per 1000). For PAs, highest rate of malpractice was 2.4 per 1000 in 2011 and lowest was 1.4 per 1000 in 2007. Most common groupings when aggregating provider groups were diagnosis related (32.2%), surgery related (26.0%), and treatment related (19.8%). PAs and NPs were significantly more likely to have diagnosis related and treatment related malpractice allegations than were physicians (each P<0.001) | Weak. Retrospective study. Confounders: not all malpractice/ adverse events are reported or claimed for; plaintiff may hold physician, as supervisor, accountable for actions of his or her employees |
| de Lusignan et al, 201666 | UK. General practice | Comparative observational study. PAs v GPs | Review of video recordings of 62 consultations (41 GP consultations and 21 PA consultations) | Patient characteristics. Number of presenting complaints. Nature of presenting complaint(s). No of relevant chronic conditions. Quality of consultation across 6 domains using Leicester Assessment Package. Inter-rater reliability | 12 GP practices were recruited, 6 with PAs and 6 without. Five GPs and 4 PAs participated. Adult patients attending for same day appointments were informed of study as they arrived and consent obtained. Assessors were experienced GPs blinded to role of clinician performing consultation. 54% of consultations were for minor symptoms/conditions. Statistically significant differences were found in presenting complaints, with GP patients having more presenting complaints per consultation (P=0.01) and more likely to have presenting complaint related to chronic condition (P=0.01) compared with PA patients. No of chronic conditions was not related to seeing GP or PA | Weak. Prospective design but small numbers. Confounders: volunteer participants, differences in patient characteristics between groups, limited inter-rater reliability of Leicester Assessment Package, unknown whether patients were triaged to PA care v GP care |
| Of rating of A-E (A being best and E being worst), average global score was C (range A-D) with no PA being given overall A for any consultation. GPs were rated more highly than PAs for all elements of consultation. In terms of median scores, this was statistically significantly higher for patient management and problem solving domains (P<0.001). Assessors were able to correctly identify GP consultations but failed to correctly identify 2/4 of PAs’. Leicester Assessment Package had limited inter-rater reliability (κ=0.602, 95% CI 0.428 to 0.777) and did not contain “complexity of consultation” rating |
CI=confidence interval; GP=general practitioner; NNB=number needed to biopsy; NP=nurse practitioner; OR=odds ratio; PA=physician assistant.
Table 5
Summary of results—patient satisfaction
| Author, year | Country and setting | Study design and comparators | Data analysed | Measures | Results | Quality assessment |
|---|---|---|---|---|---|---|
| Hooker et al, 200567 | USA. Primary care | Questionnaire survey. Older patients’ satisfaction. Physicians vNPs v PAs | Review of 146 880 completed surveys that met inclusion criteria from 321 407 randomly selected records | Patient characteristics. Patient payment source. Self-reported health. Patient satisfaction measures: attention, communication/responsiveness, respect of values/thoughts, overall rating | 2.8% (n=3770) of respondents identified PA or NP as their personal provider. Patient age groups were evenly distributed across providers. Statistically significant differences were seen in patient payment source between groups, with NPs and PAs more likely to see Medicaid enrolees (16.5% NPs v 14.1% PAs v9% physicians; P<0.001) and physicians more likely to see patients with supplemental health insurance (72.3% NPs v 76.8% PAs v 85.6% physicians; P<0.001). After applying multivariate analyses of covariance, no statistically significant difference was found for 4 satisfaction measure questions between providers | Weak. Response rate unknown. Confounders: non-responder bias; people who were unsatisfied with one provider may have moved to another leaving satisfied patients behind |
| Griffith et al, 202368 | USA. Academic dermatology centre | Questionnaire survey. PAs vresidents vdermatologists | Review of routinely collected patient satisfaction surveys following 12 386 outpatient visits. Dermatologists vresidents v PAs | Patient satisfaction measures: timeliness, patient centredness, time spent with patient, likelihood of recommending care provider | No data on patient characteristics. No of questionnaires: dermatologists 8988, residents 892, PAs 2479. Mean scores for each item for all 3 groups was high (all answers scoring >4.5/5.0). Scores were slightly lower for residents and reached statistically significant difference (P<0.01), but effect size was very small—eg, mean scores for satisfaction with time spent with patient were 4.76 for residents v 4.90 for PAs (95% CI for difference −0.17 to −0.1; Cohen’s d=0.29) | Weak. Response rate unknown. Confounders: non-responder bias. |
| Meijer and Kuilman, 201769 | Netherlands. Primary care (out of hours) | Questionnaire survey. PAs vGPs | 202 patients who had home visit out of hours and completed validated Consumer Quality Index questionnaire | Patient characteristics. Patient satisfaction measures: questions made up 3 composite subscales: approach, professional practice, and customised care plus overall satisfaction | 800 questionnaires were posted, evenly distributed to patients who had seen GP or PA. 24.8% (n=99) of patients seen by GPs and 28.8% (n=115) of those seen by PAs responded. No statistically significant difference in patient characteristics by provider type. No statistically significant difference in patient satisfaction with care provided by GP v PA for items and composite scores in professional practice and approach scales. However, two items in customised care scale, as well as related scale score, showed statistically significant difference between provider types, favouring PAs | Weak. Low response rate. Confounders: non-responder bias; owing to urgency of call, patients may not be aware which type of provider they saw |
CI=confidence interval; GP=general practitioner; NP=nurse practitioner; PA=physician assistant.
Table 6
Summary of results—cost effectiveness
| Author, year | Country and setting | Study design and comparators | Data analysed | Measures | Results | Quality assessment |
|---|---|---|---|---|---|---|
| Morgan et al, 201970 | USA. Primary care (Veterans’ Healthcare Administration) | Retrospective cohort study. Utilisation and costs of care. Physicians vNPs v PAs as primary provider | Review of 47 236 records of medically complex patients with diabetes | Patient characteristics. Patient payment source. Diagnoses. Healthcare utilisation. Costs of care | Primary provider was physician for 78.1% (n=36 894) of patients, NP for 16% (n=7536), and PA for 6% (n=2806). Patients saw their identified provider in 70.6% of visits. Differences seen in patient characteristics between groups (eg, physicians saw more black, Asian, and Hispanic patients). 65% of physician patients were in facilities with endocrinology referral capacities, compared with 58% for PAs and 52% for NPs. Similar patient payment source between groups. In terms of diagnostic score group and comorbidities, patients were similar between groups | Weak. Non-randomised retrospective design. Confounders: team based care and average practice (panel) size being higher for physicians compared with PAs not accounted for. Quality of care not assessed |
| After adjustment for differences in case mix, patients of PAs were less likely to incur hospital admissions than those of physicians (OR 0.92, 95% CI 0.846 to 0.997). Most inpatient admissions were related to ambulatory care sensitive conditions. Patients of PAs also visited ED less frequently in year than patients of PAs (PAs v physicians rate ratio: 0.94, 95% CI 0.88 to 0.99). However, when adjusted for patient characteristics, inpatient costs were not statistically significantly different between physicians and PAs. Pharmacy and outpatient costs were statistically significantly lower for PAs v physicians: 9% (95% CI 4% to 13%) lower (P<0.001) for pharmacy and 5% (1% to 9%) lower for outpatients (P<0.05). Total costs were therefore lower for PAs vphysicians by 7% per year ($2300) | ||||||
| Horak et al, 202071 | USA. Orthopaedic clinics (Veterans’ Healthcare Administration) | Questionnaire survey then economic modelling. Postoperative care, PA vsurgeon | Questionnaire of 22 clinicians followed by economic modelling | Top 5 procedures performed annually. Mean No of postoperative encounters. Mean length of postoperative visit. Annual cost of postoperative care based on PA v surgeon salaries | Questionnaires sent to 44 surgeons and 44 PAs. 25% response rate (6 surgeons and 16 PAs). Using information from survey on procedures, mean No of postoperative encounters, and mean length of postoperative visits, clinical team was found to spend 443.3 h/year on postoperative care. Using salaries of surgeons v PAs, surgeon:PA expense ratio was 5.1:1 or $122 797 per year for surgeon v $23 871 per year for PA, leading to more revenue generated per procedure owing to lower costs | Weak. Non-randomised retrospective design. Confounders: quality of care not assessed. |
| Smith et al, 202072 | USA. Adult diabetes care (Veterans’ Healthcare Administration) | Retrospective cohort study. Utilisation and costs. NPs vPAs vphysicians | Review of 368 481 records of patients with diabetes for year 2013 attending 568 Veterans’ Administration facilities | Patient characteristics. Social complexity measures. Body mass index. Global health status. Utilisation outcomes: hospital admission, ED visits, primary care visits, endocrinology outpatient visits, non-endocrinology outpatient visits. Total healthcare costs | Primary provider was physician for 74.9% (n=276 009) of patients, NP for 18.2% (n=67 120), and PA for 6.9% (n=25 352). Differences seen in patient characteristics between groups (eg, physicians saw more black, Asian, and Hispanic patients). 55.2% of physician patients were in facilities with endocrinology referral capacities, compared with 43.8% for PAs and 41.4% for NPs. Similar patient payment source between groups. In terms of diagnostic score group and comorbidities, patients were similar between groups | Weak. Non-randomised retrospective design. Confounders: team based care and average practice (panel) size being higher for physicians than PAs not accounted for. Quality of care not assessed |
| After adjustment for differences in case mix, patients of PAs were less likely to incur hospital admissions (OR 0.92, 95% CI 0.87 to 0.97) and visit ED (mean of 0.59 v 0.67 visits per year for physicians). No clinically meaningful differences were observed for No of primary care visits or endocrinology or non-endocrinology specialty visits per year. PA patients incurred lower inpatient, outpatient, pharmacy, and total costs compared with physicians, overall incurring 6% (95% CI 3% to 9%) lower cost per year, equating to $696 per year | ||||||
| Timmermans et al, 201773 | Netherlands. Inpatient medical and surgical wards | Multicentre non-randomised, matched controlled study. Cost effectiveness. Wards run by PAs >51% of time during 8 am to 6 pm weekdays vwards run by residents, both groups supervised by physicians | Review of 2292 records of adult inpatients on participating wards | Patient characteristics. Patient QALY (generic measure of disease burden). Costs of admission and within 1 month of discharge. Personnel costs | Baseline characteristics of patients differed in specialty, hospital type, major diagnostic group, type of admission (elective v urgent), and discharge destination (P<0.001). No statistically significant difference in QALY between groups (0.02, 95% CI –0.01 to 0.05). When adjusted for medical specialty, hospital type, diagnosis, comorbidities, type of admission, and discharge destination, mean total costs per patient did not differ significantly between groups: mean difference €568 (95% CI −€254 to €1391; P=0.17) | Weak. Retrospective design. Confounders: 76% response rate to questionnaire sent to patients after discharge, wards not run by PAs 100% of time, team based care and practice patterns of supervisors not accounted for |
| In sub-analyses: costs for length of stay were on average €465 (95% CI $10 to $920) per patient lower in physician model than in PA model (P=0.04). Personnel costs were lower for PA group (mean difference −€11 (95% CI −€16 to −€6) per patient (P<0.01), but this was offset by increased supervision costs for physicians (€43 (95% CI €39 to €47) per patient; P<0.01). When 4 wards staffed only by medical specialists were removed from analysis, supervision costs were significantly lower for PA wards (mean difference −€11 (95% CI −€16 to −€6; P<0.01). Significant differences in costs for blood products and required home care were higher in PA model (P<0.01) |
CI=confidence interval; ED=emergency department; GP=general practitioner; NP=nurse practitioner; OR=odds ratio; PA=physician assistant; QALY=quality adjusted life years.
Conclusion The evidence found in this review is limited and does not support the safety or effectiveness of indirect supervision of physician assistants in undifferentiated (pre-diagnosis) settings. National guidance on the supervision and scope of practice for physician assistants can ensure that physician assistants practise safely and effectively.
Study registration PROSPERO CRD42024614992.
