Factors Associated with Clinician Self-Reported Resource Use in Acute Care and Ambulatory Pediatrics

Abstract

The objective of this study is to determine predictors of resource use among pediatric providers for common respiratory illnesses. We surveyed pediatric primary care, emergency department (ED)/urgent care (UC), and hospital medicine providers at a free-standing children’s hospital system. Five clinical vignettes assessed factors impacting resource use for upper respiratory infections, bronchiolitis and pneumonia, including provider-type, practice location, tolerance to uncertainty, and medical decision-making behaviors. The response rate was 75.3% (168/223). ED/UC and primary care providers had higher vignette scores, indicating higher resource use, compared to inpatient providers; advanced practice providers (APPs) had higher vignette scores compared to physicians. In multivariate analysis, being an ED/UC provider, an APP, and greater concern for bad outcomes were associated with higher vignette scores. Overall, provider type and location of practice may predict resource use for children with respiratory illnesses. Interventions targeted at test-maximizing providers may improve quality of care and reduce resource burden.

Background

Unnecessarily high resource use and testing are common in pediatric medicine and impact patient outcomes and healthcare costs.^1,2 Unnecessary testing and treatment, defined as tests not needed to determine a diagnosis and treatments not needed for symptom and disease resolution, is frequently seen in respiratory illnesses including upper respiratory infections, bronchiolitis and pneumonia, which are among the most common illnesses evaluated by pediatric clinicians. Interestingly, unnecessary testing and treatment continues to persist despite national guidelines recommending against the use of interventions for these disease processes.^3-7 For example, the American Academy of Pediatrics recommends supportive care and against the routine use of bronchodilators, chest radiographs (chest x-ray), and viral testing for the treatment of infants with bronchiolitis.³ Unnecessarily high resource use has considerable clinical implications including exposure to invasive procedures, increased length of stay and hospitalization rates, and higher healthcare costs.^1,2

It is well known that diagnostic uncertainty is inherent in medicine and impacts clinician practice patterns and resource use. Diagnostic uncertainty exists as clinicians make decisions based on imperfect data, limited knowledge, and undifferentiated symptoms that can change over time.^8,9 Prior studies in adult primary and acute care settings have shown that diagnostic uncertainty is associated with over-testing, unnecessary procedures, and excess healthcare use and costs.^10-12 However, few studies have evaluated the impact of clinician uncertainty on diagnostic test and treatment use in the pediatric setting.

The objective of this study was to determine predictors of unnecessarily high resource use among pediatric providers for common respiratory illnesses. We specifically aimed to identify provider-level factors associated with unnecessarily high test and treatment use with the goal of informing future effective implementation/de-implementation strategies and quality improvement efforts to reduce the resource burden for these common pediatric conditions. We hypothesized that intolerance to uncertainty and perceived risk would be most predictive of unnecessarily high resource use among pediatric providers.

Methods

Study Design

This was an exploratory study conducted at a free-standing, academic, quaternary-care children’s hospital system. This study was approved by the Colorado Multiple Institutional Review Board as exempt research not requiring written consent.

Study Setting and Population

All pediatric primary care, emergency department (ED) and urgent care (UC), and hospital medicine providers who practice within the children’s hospital system were invited via email to participate in a survey from May-June 2019. We included all physicians (MD/DO) and advanced practice providers (APPs) (i.e., physician assistants (PA) and nurse practitioners (NP)). We excluded trainees including medical students, residents and fellows as their clinical decisions are impacted by decisions made by the supervising physician. The survey was administered electronically using REDCap (Research Electronic Data Capture), a HIPAA-compliant, web-based application designed by Vanderbilt University to support data collection for research studies. All study participants were informed at the beginning of the survey that completion of the survey implied consent for participation.

Survey Items and Measurements

The survey was designed to assess provider resource use for common pediatric respiratory illnesses, and provider-level factors impacting resource use including type of provider (MD/DO vs. APP), location of practice (ED/UC vs. primary care vs. inpatient), tolerance to uncertainty and medical decision-making behaviors. The survey also assessed provider demographics (i.e., age, sex, race, ethnicity), years of practice, type of degree, and location of practice (see Supplemental Material for survey).

Clinical vignettes:

The survey consisted of five clinical vignettes evaluating diagnostic, treatment, and disposition decisions for respiratory illnesses including upper respiratory infections, bronchiolitis and pneumonia. Clinical vignettes were generated by study investigators. Vignettes were then modified based on independent interviews conducted with a convenience sample of 12 providers representing each location of practice to assess their understanding of survey questions. Providers were asked to think aloud as they read and interpreted the question and answer choices. They were then asked to talk through their thought process in selecting their response, including what information they considered, and whether they felt any response options were missing. Modifications were made after each interview until respondent understanding matched what was intended by the study team. Participants also assessed survey literacy and comprehensibility, which confirmed that questions and formatting utilized were best able to elicit precise responses. The 12 providers involved in survey review and modification were excluded from the study to avoid bias.

Clinical vignettes were modified based on location of practice to align with the resources and disposition options available in the various clinical settings. Respondents recorded their likelihood of ordering specific diagnostic tests or treatments using a 5-point Likert scale. A numerical value from 1 (almost never) to 5 (almost always) was assigned to all Likert scale questions, such that a higher score indicated more testing or treatment. Study investigators created a vignette score, computed as the sum of the numeric Likert scale responses to all 16 test and treatment vignette questions on the survey, with a possible range of 16 to 80. The vignette score was used as a proxy to describe provider self-reported resource use with a higher score indicating higher resource use. One clinical vignette posed a series of questions to assess providers’ pre- and post-test perception of the probability of pneumonia, from 0% to 100%, in an infant with bronchiolitis given additional diagnostic findings (i.e., viral testing and chest x-ray). Of note, successive questions with additional diagnostic findings were only presented after survey participants completed prior questions to limit biasing responses towards higher resource use.

Provider perception of risk and intolerance:

For assessment of tolerance to uncertainty and medical decision-making behaviors, the we adapted two previously validated tools, the Physician Reaction to Uncertainty (PRU) scale and Medical Maximizer-Minimizer Scale modified for providers (MMS).^13,14 The adapted PRU scale consists of 15-items, assessed using a 6-point Likert scale (1-strongly disagree, 6-strongly agree) and divided into four components: 1) anxiety due to uncertainty (5-items), 2) concern about bad outcomes (3-items), 3) reluctance to disclose uncertainty to patients (5-items), and 4) reluctance to disclose mistakes to physician colleagues (2-items). The original 10-item MMS was developed and validated to assess patient medical minimizing and maximizing behaviors and used a 7-point Likert scale.¹⁴ The adapted MMS, used in this study, was created to assess physician and APP medical decision-making behaviors. The adapted MMS has been used in prior research but has not been previously validated.¹⁵ In contrast to the original 10-item MMS, the adapted MMS consists of 8-items: 1 clinician-centric item was added, and 3 patient-centric items were removed. The adapted MMS also uses a 6-point Likert scale (1-stongly disagree, 6-strongly agree), similar to the previously validated single-item maximizer-minimizer elicitation question (the MM1).¹⁶ (See Appendix for survey)

Data Analysis

Descriptive statistics were utilized to describe provider demographics and summarize survey responses. Continuous variables were described using means (±SD) and medians (IQR), and categorical variables were described using frequencies and proportions. Chi-square, Fisher’s exact tests and t-tests were used to assess differences in categorical and continuous respondent characteristics by practice location type. Responses to the individual clinical vignettes were dichotomized as “almost always/often” and “sometimes/rarely/almost never”, and chi-square tests were used to assess differences in the probability of ordering tests/treatments by practice location and provider type. Student’s t-tests were used to compare the probability of diagnosing pneumonia by providers who “almost always/often” order tests and treatments (e.g., chest x-ray and antibiotics) compared to those who “sometimes/rarely/almost never” order tests and treatments. Chi-square tests were used to compare the distribution of those reporting 75-100% pre-test probability of pneumonia by provider type.

As recommended by authors of the PRU and MMS scales, scale responses were averaged for each category (e.g., “anxiety of uncertainty”).^13,14 Cronbach’s alpha was computed to assess the reliability of the MMS as this scale was not previously validated to describe provider medical decision-making behaviors. There are no official standards for defining good or acceptable Cronbach’s alpha, however the literature has generally accepted alpha values >.70.¹⁷ Student’s t-tests were used to compare mean PRU, MMS, and vignette scores by type of provider and analysis of variance was used to compare mean scores by location of practice.

Multivariable regression analyses were performed to determine the association between provider-level factors and resource use as measured by the total vignette score. Statistical significance was assessed at the 0.05 significance level and 95% confidence intervals were reported. Statistical analysis was performed using SAS software (ver. 9.4, SAS Institute, Cary, NC).

Results

Survey response and respondent characteristics

The final sample size was 168 with an overall response rate of 75.3% (168/223). The majority of survey respondents were female (78.5 %), physicians (65%), non-Hispanic (94.0%) and had ≥ 5 years of experience since completion of training (70.5%) (Table 1). The distribution of primary location of practice of providers was primary care clinic (15.5%), ED/UC (58.9%) and inpatient (25.6%). There were no significant differences in respondent characteristics by primary location of practice.

Table 1.

Demographics of respondents by primary location of practice.

Variables	Overall N=166	Primary Care N=26	ED/UC N=97	Inpatient N=43	P-value
Demographics
Age, mean±SD*	40.1±8.0	44.2±9.7	39.8±7.5	38.3±7.1	0.60
Sex (female)	128 (78.5)	20 (76.9)	78 (80.4)	30 (69.8)	0.26
Race (White) †	144 (86.8)	23 (88.5)	83 (85.6)	38 (88.4)	0.72
Ethnicity (non-Hispanic) †	156 (94.0)	24 (92.3)	91 (93.8)	41 (95.4)	0.75
Provider characteristics
Provider type (MD/DO)	108 (65.1)	17 (65.4)	58 (59.8)	33 (76.7)	0.15
Time in direct patient care (≥ 75%)	88 (53.0)	6 (23.1)	60 (61.9)	22 (51.2)	<0.001
Years in practice (1-4 years)	49 (29.5)	3 (11.5)	29 (29.9)	17 (39.5)	0.05

All comparisons performed with chi-square tests unless otherwise specified.

^*T-test

^†Fisher’s exact test

Clinical vignettes

The total clinical vignette score ranged from 22 to 56 with a mean of 38.7±7.1 and median 38.5 (IQR 34, 43). There were statistically significant differences in the vignette score by location of practice (p<0.001), with ED/UC and primary care providers having higher vignette scores than inpatient providers. Similarly, APPs had higher vignette scores compared to physicians (p=0.002) (Table 2).

Table 2.

Physician uncertainty measures by provider type and location of practice.

Score	Provider Type			Location of practice
Variable	MD/DO N=108	APP N=58	P-value	Primary Care N=26	ED/UC N=97	Inpatient N=43	P-value
Vignette Score	37.4±6.6	41.1±7.5	0.002	38.9±8.4	40.5±6.5	34.5±5.8	<0.001
Gerrity- Physician Reaction to Uncertainty (PRU) Score
Anxiety due to uncertainty score	15.7±5.2	18.1±4.4	0.006	15.4±4.5	16.5±5.2	17.3±5.2	0.36
Concern about bad outcomes	9.7±3.6	10.4±3.3	0.21	9.0±3.8	10.2±3.2	9.9±3.8	0.33
Reluctance to disclose uncertainty to patients	1 2.5±3.7	14.0±4.5	0.03	11.2±3.3	13.6±4.2	12.8±3.8	0.03
Reluctance to disclose mistakes to physicians	4.2±2.1	4.1±2.4	0.83	4.1±2.0	4.3±2.4	4.1±2.0	0.90
Medical Minimizer/Maximizer Score	2.3±0.7	2.5±0.6	0.08	2.2±0.7	2.4±0.7	2.2±0.6	0.23

^*Results presented as mean score ± standard deviation.

For the individual clinical vignettes, several test and treatment decisions demonstrated statistically significant differences by location of practice and provider type. In the case of a 6-week old with bronchiolitis, ED/UC and primary care providers were more likely to report almost always/often ordering a viral test (ED/UC 16.7%, primary care 24%, inpatient 2.4%; p=0.02) or chest x-ray (ED/UC 28.1%, primary care 28%, inpatient 11.9%; p=0.01) compared to inpatient providers. Similarly, APPs were more likely to report almost always/often ordering a viral test (APP 22.8% vs. MD/DO 9.4%; p=0.005) or chest x-ray (APP 35.1% vs. MD/DO 17.9%; p=0.004) compared to physicians for the same clinical scenario.

In the case of an 8-week old with bronchiolitis, ED/UC and primary care providers were more likely to report almost always/often ordering a catheterized urinalysis compared to inpatient providers (ED/UC 61.3%, primary care 33.3%, inpatient 26.8%; p <0.001). There was no difference between physicians and APPs in obtaining a catheterized urinalysis in this clinical scenario. However, APPs were more likely to report almost always/often obtaining complete blood count and blood cultures for the same patient APP 32.7% vs. MD/DO 11.7%; p<0.001).

For the clinical vignette assessing the probability of bacterial pneumonia in an infant with a history and exam consistent with viral bronchiolitis, there was a trend towards APPs reporting a higher pre-test probability of diagnosing bacterial pneumonia prior to test results; 73.3% of APPs compared to 26.7% of physicians reported a 75-100% pre-test probability for diagnosing pneumonia (p=0.07). Overall, APPs reported a higher probability of diagnosing pneumonia based on history, physical exam and additional diagnostic findings including positive viral panel results (Figure 1). APPs were also more likely to order tests and treatments (i.e., chest x-ray, viral testing, antibiotics) and less likely to discontinue treatments (i.e., antibiotics) when given additional findings (Figure 2). In addition, providers who were more likely to order chest x-rays reported a higher pre-test probability for diagnosing pneumonia compared to those who were unlikely to order chest x-rays (50.3% vs. 22%; p<0.001). Subsequently, those who diagnosed pneumonia based on chest x-ray results were also more likely to order antibiotics (p<0.001).

Figure 1. Change in provider report of probability of bacterial pneumonia given additional diagnostic findings by provider type (MD/DO vs. APP).

Provider-reported probability of bacterial pneumonia based on:

1) History and physical exam alone.

2) Addition of chest-x-ray showing airways disease and right middle lobe infiltrate.

3) Addition of viral panel positive for Respiratory Syncytial Virus.

Clinical vignette for Figures 1 and 2: In January, a 6-week old girl with rhinorrhea and three days of cough presents to your office. Per parents, she began breathing rapidly last evening. Slightly decreased feeding today but plenty of wet diapers. She was born full term and is up-to-date on immunizations without any medical problems. She is alert with a temperature of 38.3, respiratory rate 50, and 87% oxygen saturation on room air. She appears well-hydrated and non-toxic, but has had mild retractions, scattered expiratory wheezing and asymmetrical exam (fine crackles right > left).

Figure 2. Likelihood of ordering or discontinuing tests or treatments given additional diagnostic findings by provider type (MD/DO vs. APP).

Clinical vignette for Figures 1 and 2: In January, a 6-week old girl with rhinorrhea and three days of cough presents to your office. Per parents, she began breathing rapidly last evening. Slightly decreased feeding today but plenty of wet diapers. She was born full term and is up-to-date on immunizations without any medical problems. She is alert with a temperature of 38.3, respiratory rate 50, and 87% oxygen saturation on room air. She appears well-hydrated and non-toxic, but has had mild retractions, scattered expiratory wheezing and asymmetrical exam (fine crackles right > left).

PRU and MMS scores

The average summed scores for subcategories of the PRU were: anxiety due to uncertainty (16.5±5.1), concern about bad outcomes (9.9±3.5), reluctance to disclose uncertainty to patients (13±4.1) and reluctance to disclose mistakes to physicians (4.2±2.2). The average summed MMS score was 2.3±0.7. The Cronbach’s alpha for the MMS was 0.79, demonstrating acceptable reliability to assess provider medical minimizing and maximizing behaviors. There were no statistically significant differences in PRU or MMS scores by location of practice, with the exception of “reluctance to disclose uncertainty to patients” (ED/UC: 13.6±4.2, inpatient: 12.8±3.8, outpatient: 11.2±3.3; p=0.03) (Table 2). However, APPs had higher “anxiety due to uncertainty” (p=0.006) and “reluctance to disclose uncertainty to patients” (p=0.03) scores compared to physicians. There was no difference in MMS scores by provider type (MD/DO: 2.3±0.7, APP: 2.5±0.6; p=0.08).

Predictors of test and treatment utilization

A multivariable regression was performed to identify which set of provider-level factors best predict resource use as measured by the clinical vignette score. Included in the model were primary location of practice, provider type, age of provider, sex of provider, years in practice (< 5 years vs. >= 5 years), the 4 subscales of the PRU (1) anxiety due to uncertainty, 2) concern about bad outcomes, 3) reluctance to disclose uncertainty to patients, and 4) reluctance to disclose mistakes to physician colleagues), and the MMS. Being an ED/UC provider, being an APP, and greater concern about bad outcomes were significantly associated with higher vignette scores (Table 3).

Table 3.

Multivariable regression model for provider-level of resource use as measured by vignette scores.

Variable	Estimate	Confidence Interval	P-value
Practice location (ref: ED/UC)
Primary care	0.016	−3.19 - 3.22	0.99
Inpatient	−5.182	−7.71 - −2.65	<.001
Provider type (ref: APP)
MD/DO	−2.662	−0.33 - −4.99	0.03
Age	0.029	−0.14 - 0.20	0.74
Sex (ref: female )
Male	1.162	−1.59 - 3.91	0.40
Years in practice (ref: <5 years)
>=5 years	0.352	−2.37 - 3.08	0.80
PRU subscale scores
Anxiety due to uncertainty	−0.031	−0.32 - 0.26	0.84
Concern about bad outcomes	0.416	0.02 - 0.81	0.04
Reluctance to disclose uncertainty to patients	0.176	−0.13 - 0.48	0.26
Reluctance to disclose mistakes to physician colleagues	0.141	−0.37 - 0.65	0.59
MMS (Medical Minimizer-Maximizer Scale)	1.258	−0.48 – 3.00	0.15

^*PRU subscale and MMS scale scores included in the model are averaged summed scores across all provider types.

Discussion

Our study of pediatric clinicians suggests that resource use is informed by provider type and location of practice rather than provider degree of uncertainty or perceived risk. Our results demonstrate that being an ED/UC provider, being an APP, and having concern for bad outcomes are most predictive of higher resource use. Thus, contrary to our hypothesis, implementation/de-implementation strategies and quality improvement efforts might be more effective if targeted at specific provider types and locations of practice, rather than uncertainty scores.

Vignette scores were higher among outpatient providers (i.e., ED/UC and primary care providers) compared to inpatient providers. Interestingly, there was no difference in uncertainty and medical maximizer-minimizer scores by location of practice. The higher resource use noted in the outpatient setting is likely a result of the different goals of practice in these settings. ED/UC and primary care providers have a short period to evaluate patients and develop a diagnostic decision and treatment plan. Thus, these providers may desire more diagnostic certainty prior to discharging patients as follow-up is not definitive and there is a fear of missed or delayed diagnoses. Contrary to the outpatient setting, inpatient providers have the opportunity to develop a rapport with caregivers and observe patients for longer periods, and thus can make diagnostic and treatment decisions based on more data and symptoms that declare themselves over time. In addition, inpatient settings have been quicker to adopt strategies to promote high-value care, which likely contributes to inpatient providers being stewards of lower resource use.^18-20 Lastly, providers were not surveyed on their tolerance or concern for medical legal risk, which may impact test and treatment use, especially in the ED/UC setting.

As a group, APPs reported higher resource use compared to physicians. This variance in practice by provider type could be explained by uncertainty scores as APPs had higher uncertainty scores in two domains compared to physicians (i.e., “anxiety due to uncertainty” and “reluctance to disclose uncertainty to patient”). APPs may order more tests and treatments to provide reassurance to patients and caregivers given their greater reluctance to disclose uncertainty. Other studies have found similar differences in resource utilization between APPs and physicians; APPs ordered more imaging studies,^21,22 made more specialist referrals for patients with diabetes,²³ and were more likely to prescribe antibiotics for acute upper respiratory infections compared with physicians.²⁴

The higher resource use and higher degree of uncertainty reported by APPs may be due in part to the different type and length of the training they receive. Providers draw on both formal training and the sum of past patient encounters when caring for patients. Clinical experience varies among APPs. Many NPs have clinical experience as bedside nurses prior to obtaining their advanced degrees, but this is not required. Residency training for physicians, by contrast, is more standardized and designed to expose trainees to a wide range of clinical presentations prior to independent practice. In addition, there is substantial variability in the scope of practice of APPs. For example, while NPs mostly practice independently, PAs require physician supervision in 43 states, including Colorado.²² At our institution, APPs also care for patients of a variety of acuity levels. In some ED/UCs, APPs see low acuity patients defined by Emergency Severity Index (ESI) 3-5, but at others APPs may care for higher acuity patients (ESI 1-2).²⁵ This variation in scope of practice may also influence diagnostic certainty and resource use.

In general, APPs offer an important opportunity to fill an ever-increasing need for health care providers and to lower the cost of delivering healthcare. However, our study highlights the need to improve APP knowledge of high-value care to impact health care costs. More recently, physician training programs have incorporated a curriculum to educate trainees on the importance of high-value care.^26,27 Similarly, there has been rise in APP post-graduate fellowship programs aimed at improving APPs’ knowledge and skill set necessary to provide high quality patient care and improve diagnostic uncertainty. A one-year pediatric inpatient APP fellowship program succeeded in improving APP clinical preparedness by 25%.²⁸ Such programs may improve APP clinical knowledge and certainty, and ultimately improve patient care and reduce health care costs.

In our study, providers who estimated a higher pre-test probability for diagnosing bacterial pneumonia were more likely to order chest x-rays and viral testing. This is not surprising given the recognition that history and clinical examination findings are imperfect for the diagnosis of community acquired pneumonia or to differentiate bacterial from viral etiologies.^29,30 Importantly, the incidence of bacterial pneumonia is quite low in comparison to viral bronchiolitis in young children.³¹ It is unclear from our findings whether providers order imaging in order to decrease the need for antibiotic prescriptions or to confirm their clinical suspicion for a diagnosis of bacterial pneumonia. Knowledge of disease incidence is essential for determining the potential benefit of diagnostic studies and for interpreting and acting on study results. Interestingly, APPs reported a significantly higher pre-test probability of diagnosing bacterial pneumonia and were more likely to order chest x-rays and viral testing than physicians. APPs’ also reported higher post-test probability of diagnosing bacterial pneumonia, which aligned with their pre-test probability inciting a treatment cascade leading to prescription for antibiotics and lack of discontinuation of antibiotics after knowledge of positive viral test results. All providers may benefit from more rigorous training in clinical epidemiology to understand the importance of incorporating disease incidence into clinical decision-making. Additionally, including information about disease incidence in institutional care pathways may reduce providers’ pretest probability for less common pediatric diseases thereby reducing resource utilization.

Our data suggests that the PRU and MMS were not predictive of resource use among pediatric providers as initially hypothesized. This finding is contrary to existing literature demonstrating that provider degree of uncertainty impacts resource use and costs. In a Medicare health maintenance organization (HMO), high “anxiety due to uncertainty” was associated with higher patient charges; every standard deviation increase in physician “anxiety due to uncertainty” was associated with a 17% increase in mean patient charges.¹⁰ Similarly, in a study of radiologists evaluating diagnostic mammograms, higher physician uncertainty scores were associated with higher recall rates, lower specificity and lower positive predictive value in diagnostic mammography interpretation.¹¹ One possibility for the finding in our study is that providers did not report their true degree of uncertainty or perceived discomfort with risk. In addition, given most survey respondents had ≥ 5 years of experience, they may have developed strategies for coping with uncertainty decreasing the impact uncertainty has on medical decision-making. Furthermore, recent focus on provider resilience and burn-out reduction may have resulted in improved management of clinical uncertainty. Lastly, the MMS has not been validated to assess provider medical decision-making behaviors; thus, may not have truly captured provider risk behaviors. However, our reliability analysis of MMS items demonstrated reasonable reliability among items.

Overall, our findings highlight opportunities for implementation/de-implementation strategies and quality improvement efforts to decrease unnecessary resource use among pediatric providers. Strategies might be more effective if focused on provider-types (i.e., APPs) and locations of practice (i.e., ED/UC and primary care) that demonstrated higher resource use. Such strategies include incorporation of high-value care curricula into training programs, post-graduate fellowship programs for APPs, use of standardized clinical pathways for common pediatric illnesses, and provider feedback.^18,20,27 These strategies, along with on-going directed continuing medical education, have potential to improve patient care and outcomes by improving provider understanding of the clinical, quality and financial drivers that impact patient outcomes.

This study has several limitations. First, it is an exploratory, survey-based study. Thus, our data are prone to subjectivity and response bias. We relied solely on provider report of clinical practice, which may not reflect actual clinical decisions. However, our response rate captured most primary care, ED/UC and hospital medicine providers at our institution, thus limiting sampling bias. In addition, our analysis did not delineate between APP type (i.e., NPs and PAs). APPs are a diverse group with a range of training and clinical experience, and we did not evaluate for differences between APP type, breadth of clinical experience or post-graduate training, which all impact resource use. Furthermore, the adapted MMS used for this study was not previously validated to assess provider medical decision-making behaviors. However, we performed a reliability analysis demonstrating that the items of the MMS had reasonable reliability to assess provider medical decision-making behaviors. In addition, we were unable to control for all predictors of treatment use, and the clinical vignette score used to describe provider level of resource use has not been previously validated. Lastly, this study surveyed providers at multiple sites within a single institution and may not be generalizable across other settings.

Conclusion

High resource use continues to persist for common pediatric respiratory illnesses despite lack of evidence to support diagnostic testing and treatment for these conditions. Our study suggests that implementation/de-implementation strategies and quality improvement efforts should be targeted at specific provider types and locations of practice. Such efforts will improve the quality of care and clinical outcomes for children by advancing towards high-value care.