Facility variation in troponin ordering within the Veterans Health Administration

Abstract

Background

Current United States guidelines recommend troponin as the preferred biomarker in assessing for acute coronary syndrome, but recommendations are limited about which patients to test. Variations in troponin ordering may influence downstream healthcare utilization.

Methods

We performed a cross-sectional analysis of 3,308,131 emergency department (ED) visits in all 121 acute care facilities within the Veterans Health Administration (VHA) from 2015–2017. We quantified the degree to which case mix and facility characteristics accounted for variations in facility rates in troponin ordering. We then assessed the association between facility quartiles of risk-adjusted troponin ordering and downstream resource utilization (inpatient admissions, non-invasive testing [stress tests, echocardiograms], and invasive procedures [coronary angiograms, percutaneous coronary interventions, and coronary artery bypass grafting surgeries].

Results

The proportion of ED visits with troponin orders ranged from 2.2% to 64.5%, with a median of 37.1%. Case mix accounted for 9.5% of the variation in troponin orders; case mix and differences in facility characteristics accounted for 34.6%. Facilities in the highest quartile of troponin ordering, as compared to those in the lowest quartile, had significantly higher rates of inpatient admissions, stress tests, echocardiograms, coronary angiograms, and PCI.

Conclusions

Significant variation in troponin utilization exists across VHA facilities and that variation is not well explained by case mix alone. Facilities with higher rates of troponin ordering were associated with more downstream resource utilization.

Introduction

Patients in the emergency department (ED) with suspected acute coronary syndrome (ACS) account for more than seven million visits and over $5 billion in direct healthcare expenditures annually [1–2]. Chest pain patients include patients both at very high and low risk for ACS [3–4]. Cardiac troponin is a biological marker of myocardial injury that is often used to diagnose or exclude cases of acute myocardial infarction (AMI). Medical practitioners have accordingly adopted serial troponin measurements as an essential step in evaluating patients with suspected ACS [5]. However, current guidelines do not provide specific recommendations concerning appropriate diagnostic testing thresholds for measuring troponins in this clinically heterogenous population [6–8]. Differences in troponin testing practices therefore may not only reflect variations in case mix but also reflect under or overutilization of troponin testing [9].

At present, there is little information on interfacility variation of troponin ordering. Underutilization of troponin testing could lead to delayed diagnoses of ACS, while liberal testing may lead to detection of higher rates of elevated troponin not representing AMI [10–11]. These “false positive” results may in turn lead to expensive and invasive downstream healthcare utilization including hospital admissions, non-invasive diagnostic testing, and invasive procedures. Understanding troponin testing patterns may create opportunities to minimize excess healthcare expenditures and prevent patients from undergoing unnecessary medical procedures. Furthermore, evidence of significant variation across facilities may prompt further research into why such differences exist and spur efforts to standardize troponin testing to optimize resource utilization.

To address this gap in knowledge, we leveraged data from the nation’s largest integrated healthcare system, the Veterans Health Administration (VHA), to examine variations in troponin use in EDs across all acute care facilities. We also assessed whether variations in troponin ordering were associated with differences in downstream healthcare utilization.

Methods

Data Sources

We performed a cross-sectional analysis of all acute care facilities with an ED within the VHA from January 1, 2015 to December 31, 2017. Data for this study were obtained from the VHA’s Corporate Data Warehouse, a repository of patient-level data including demographic, clinical data, vital signs, laboratory results, and mortality data [12].

Study population

We included ED encounters at any acute care VHA facility within our study period. We excluded any encounters without laboratory orders (53% of ED encounters) as they largely represented visits where patients left without being seen by a provider or had low acuity presentations, a clinically distinct patient population than one where troponin ordering is considered. Complete blood count and metabolic panels were used as surrogate markers of laboratory orders given the expected high frequency of these tests being ordered in the ED. The proportion of encounters with no laboratory orders was consistent across facilities over the study period. We also excluded visits with possible outlier clinical data values (lower 1% and upper 99%) and those without a gender designation (<10,000). Characteristics of patient visits excluded by outlier data values are provided in the Supplementary Materials (Table 1)

Our initial cohort included 7,630,529 ED encounters. Of these, 3,656,525 had a laboratory order. After removal of encounters with possible outlier clinical data values, the final cohort included 3,308,031 encounters. Since our cohort was comprised of ED visits, individual patients could contribute to multiple observations in the data set.

Outcomes

The primary outcome was one or more troponin results available within twenty-four hours after ED arrival. Our primary outcome included both troponin T and I results. Secondary outcomes included visits with creatinine kinase-muscle/brain (CK-MB) and elevated or abnormal troponin results. We designated the encounter as associated with an abnormal troponin result if at least one of the result values within twenty-four hours of ED arrival was above the facility’s reference range. We included these secondary outcomes to determine if variations in troponin use explained by differences in CK-MB ordering or selectivity in troponin ordering.

Downstream resource utilization was defined as subsequent inpatient admissions, echocardiograms, stress tests, coronary angiograms, PCI, and CABG. We included inpatient admissions if they occurred within 24 hours after the index ED visit time. Non-invasive testing and invasive procedures were considered as outcomes if they occurred within one month of the index ED visit date and were captured using Current Procedural Terminology (CPT) codes (list of CPT codes utilized in table 2 of Supplementary Materials). We tracked patient transfers to different VHA facilities using unique patient identifiers, and we linked all downstream resource utilization with the initial ED encounter.

Study variables

We considered patient-level and facility-level characteristics for risk adjustment. Patient-level characteristics included demographics (age, sex, race); medical history (hypertension, hyperlipidemia, coronary artery disease, atrial fibrillation, congestive heart failure, diabetes, cerebrovascular accident, anemia, cirrhosis, end-stage renal disease, chronic lung disease, peripheral vascular disease, sleep apnea, ventricular arrhythmias); prior procedural history (stress test, angiogram, PCI, CABG); and clinical values (heart rate, systolic blood pressure, diastolic blood pressure, body mass index, hemoglobin, white blood count, creatinine). Given the relatively higher rates of substance use and psychiatric comorbidities in the Veteran population compared to the general population [13–15], we included the following mental health diagnoses for adjustment: alcohol dependence, illicit drug dependence, depression, schizophrenia, bipolar disorder, and post-traumatic stress disorder. For each ED encounter, we identified the patient’s medical and psychiatric comorbidities using International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) and International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) codes. We utilized both sets of codes as conversion of ICD-9 diagnostic codes to ICD-10 was not fully complete within VHA’s electronic health record during this study period.

Facility-level characteristics in this analysis included geographic region, urban/rural designation, availability of PCI, and VHA facility complexity. Facility complexity is a VHA model that designates each VHA facility into one of five distinct levels (1A, 1B, 1C, 2, and 3). The categorization is determined by patient population characteristics, offered clinical services, facility size, facility capability (e.g., number of intensive care units), and academic (research and teaching) mission [16–20].

Statistical analyses

We calculated descriptive statistics for encounter and facility level covariates, including means and standard deviations (SD); medians; ranges and interquartile ranges (IQR); and frequencies and proportions.

We divided facilities into quartiles based on the proportion of ED encounters with a troponin. Quartile 1 had the lowest rates of troponin utilization and quartile 4 the highest. We estimated Pearson’s correlation between troponin and CK-MB. Similarly, we estimated the correlation between unadjusted facility-level troponin use and the proportion of troponin results that were abnormal.

We calculated risk-standardized troponin ordering rates for each facility using hierarchical generalized linear models that account for clustering of encounters within facilities. We modeled the log-odds of troponin use as a function of demographic, clinical, and facility variables and a random hospital-specific effect. This approach allows the discrimination of between-hospital variation from within-hospital variation and case mix. These rates are the ratio of predicted to expected troponin ordering, multiplied by the mean unadjusted overall troponin rate scaled to 100 [21].

We estimated the predicted number of troponins ordered for each facility using its patient case mix and facility-specific intercept. We estimated the expected number of troponins ordered using the same patient case-mix and the average facility-specific intercept of all facilities in our analysis. This approach is consistent with hospital profiling by Medicare and is a form of indirect standardization [22].

We then constructed a sequence of hierarchical models to investigate the degree to which patient case-mix and facility characteristics contributed to variations in troponin use. For each model, we quantified the extent of variation in troponin use by calculating the median odds ratio (MOR) [23]. In this setting, the MOR is the relative odds that two identical patients will receive a troponin order at one hospital in comparison to another randomly selected hospital. In the first model, we fit an unconditional model with only a facility random intercept to determine if any between-facility variation existed. In the second model, we included demographic, medical and psychiatric history, and clinical information to quantify the amount of variation that can be explained by patient case-mix. Finally, in the third model, we also included facility characteristics to determine the variation attributable to differences in facility characteristics.

We then partitioned facilities into quartiles based on risk-standardized troponin rates. We assessed the association between quartiles of risk-adjusted troponin ordering and downstream resource utilization (inpatient admissions, non-invasive testing, and invasive procedures) using Kruskal-Wallis tests to determine if there were differences in distribution across the quartiles.

Falsification and Sensitivity Analyses

To account for potential residual confounding, we conducted falsification analysis using kidneys, ureter, and bladder (KUB) X-rays and right upper quadrant ultrasounds, as troponin ordering is unlikely to be associated with these studies [24]. Additionally, our final data set had a limited amount of missing data (<1%), but given that these data were most likely not missing at random, we employed a missing indicator method to include all observations. We performed sensitivity analyses by repeating our analyses on the entire cohort including patients without laboratory orders and only in encounters seen at level 1A, 1B, or 1C facilities. To determine if observed associations between troponin ordering and downstream utilization could be solely explained by troponin positivity rates, we repeated this analysis by stratifying the facility quartiles based on the proportion of troponin tests that resulted critically abnormal (three times above the upper limit of normal).

All analyses were performed in SAS version 9.4 (SAS Institute, Cary, North Carolina) and R version 3.5.1 (https://www.r-project.org/). We evaluated the null hypothesis with a two-sided significance level of 0.05. The West Haven VA review board approved this study and waived the requirement for informed consent.

Results

Our analysis included 121 facilities, and 38.9% of all ED encounters were associated with troponin results. Troponin rates across facilities varied widely from 2.2% to 64.5%. After partitioning the facilities into quartiles, the median troponin rates were 26.7% for quartile 1 (lowest troponin utilization), 35.1% for quartile 2, 40.8% for quartile 3, and 48.9% for quartile 4 (highest troponin utilization).

Patient and facility-level characteristics

The mean (SD) age of patients in this analysis was 62.1 (15.1), 8.6% were female, and 68.8% were white. As compared to patients in high troponin utilization facilities, patients in ED encounters from low troponin utilization facilities were younger and less likely to have hypertension, coronary artery disease, atrial fibrillation, congestive heart failure, cerebrovascular disease, anemia, cirrhosis, peripheral vascular disease, substance dependence, schizophrenia, bipolar disorder, and alcohol dependence. Patients in ED encounters from low utilization facilities were also less likely to have undergone prior stress tests, angiograms, and PCIs as compared to high utilization facilities (Table 1).

Table 1:

Baseline demographic and clinical characteristics stratified by quartiles of facility troponin ordering

Variable	Overall (N=3,199,324)	Quartile 1 (N=658,087)	Quartile 2 (N=725,483)	Quartile 3 (N=885,379)	Quartile 4 (N=930,375)
Demographics
Age, Mean (SD)	62.0 (15.1)	61.4 (15.0)	62.4 (15.6)	62.2 (14.9)	62.1 (15.0)
Race
White	68.7%	68.0%	70.7%	70.1%	66.4%
Black	23.8%	24.3%	22.1%	22.4%	25.9%
Other	7.5%	7.7%	7.2%	7.5%	7.7%
Gender
Male	91.3%	91.0%	91.5%	91.5%	91.2%
Female	8.7%	9.0%	8.5%	8.5%	8.8%
Smoking Status
Current	35.3%	34.0%	36.5%	35.2%	35.4%
Former	19.0%	20.0%	19.5%	16.7%	19.9%
Never	45.7%	46.0%	44.0%	48.0%	44.7%
Medical History
Hypertension	75.5%	74.1%	74.8%	76.7%	75.9%
Hyperlipidemia	70.6%	70.2%	70.3%	71.5%	70.3%
Coronary artery disease	30.7%	28.6%	29.7%	32.4%	31.4%
Atrial fibrillation	16.3%	14.8%	16.7%	16.7%	16.8%
Congestive heart failure	20.9%	18.6%	20.8%	22.5%	21.1%
Diabetes	41.1%	40.6%	40.8%	41.8%	41.0%
Cerebrovascular accident	16.5%	15.5%	16.5%	16.3%	17.4%
Anemia	15.5%	14.4%	15.7%	15.6%	16.1%
Cirrhosis	18.7%	17.2%	19.7%	18.5%	19.3%
End-stage renal disease	3.6%	3.1%	3.9%	3.6%	3.8%
Chronic lung disease	48.9%	48.0%	47.9%	49.9%	49.3%
Peripheral vascular disease	22.8%	21.8%	23.4%	22.8%	23.1%
Ventricular arrythmias	3.9%	3.4%	3.8%	4.2%	3.9%
Sleep apnea	29.7%	29.5%	30.8%	30.4%	28.4%
Alcohol use disorder	16.2%	15.1%	17.4%	15.9%	16.1%
Substance use disorder	30.8%	27.7%	33.9%	30.1%	31.2%
Depression	53.8%	52.6%	55.2%	53.1%	54.3%
Post-traumatic stress disorder	31.6%	30.8%	32.3%	31.1%	31.9%
Schizophrenia	8.0%	7.1%	9.6%	7.5%	7.9%
Bipolar Disorder	37.3%	36.1%	39.5%	36.9%	36.6%
Prior procedural history
Stress Test	43.1%	38.9%	43.7%	44.9%	43.9%
Angiogram	14.9%	12.7%	13.3%	17.0%	15.7%
Percutaneous coronary intervention	4.3%	3.8%	3.6%	5.1%	4.3%
Coronary artery bypass grafting surgery	1.9%	1.8%	2.0%	2.1%	1.8%
Clinical values
Heart Rate	83.5 (16.5)	83.4 (16.4)	83.1 (16.4)	83.9 (16.7)	83.6 (16.6)
Diastolic blood pressure, Mean (SD)	80.5 (13.1)	80.3 (12.9)	80.1 (12.8)	81.1 (13.4)	80.4 (13.3)
Systolic blood pressure, Mean (SD)	138.5 (22.1)	138.8 (22.0)	137.9 (21.9)	139.0 (22.3)	138.3 (22.1)
Body mass index, Mean (SD)	30.5 (6.4)	30.7 (6.4)	30.3 (6.3)	30.6 (6.5)	30.5 (6.4)
White blood cell, Mean (SD), 103/mL	8.4 (3.1)	8.4 (3.1)	8.4 (3.1)	8.5 (3.2)	8.4 (3.1)
Hemoglobin	13.6 (2.0)	13.6 (1.9)	12.5 (2.0)	13.6 (2.0)	13.5 (2.0)
Creatinine, Mean (SD), mg/dL	1.2 (0.7)	1.2 (0.7)	1.2 (0.7)	1.2 (0.7)	1.2 (0.7)

Abbreviations: SD – standard deviation

High utilization facilities were rated higher on the VHA facility complexity model when compared to lower utilization facilities (71.0% vs 46.6% rated as 1A, 1B, or 1C). High utilization facilities were more likely to be in an urban area and have PCI capabilities (Table 2).

Table 2:

Facility characteristics stratified by quartiles of facility troponin ordering

Variable	Quartile 1 (n=30)	Quartile 2 (n=30)	Quartile 3 (n=30)	Quartile 4 (n=31)
Facility Complexity
1A	23.3%	40.0%	30.0%	35.5%
1B	13.3%	16.7%	10.0%	22.6%
1C	10.0%	6.7%	30.0%	16.1%
2	20.0%	20.0%	23.3%	16.1%
3	33.3%	16.7%	6.7%	9.7%
Geographic Region
Northeast	16.7%	26.7%	6.7%	19.4%
Midwest	26.7%	20.0%	26.7%	16.1%
South	40.0%	23.3%	46.7%	45.2%
West	16.7%	26.7%	20.0%	19.4%
Puerto Rico	0%	3.3%	0%	0%
Urban (vs rural) Designation	86.7%	83.3%	90.0%	96.8%
PCI capabilities	53.3%	73.3%	83.3%	77.4%

Abbreviations: PCI – percutaneous coronary intervention

CK-MB and abnormal troponin proportions

The median rate of CK-MB ordering in this analysis was 4.9% (IQR: 0.43–17.6%). There was low-moderate correlation between facility-level troponin and CK-MB (Pearson=0.32; p<0.001). The median proportion of troponins that were elevated or abnormal was 13.9% (IQR 8.17–23.9%). There was low-moderate correlation between facility-level troponin and the proportion of abnormal troponin results (Pearson=0.49; p<0.001).

Variation in rates of troponin ordering

After adjusting for case-mix and facility characteristics, there was still significant variation across facilities in rates of troponin use (Figure 1).

Figure 1:

Individual facilities and their risk-adjusted troponin ordering rates

Each data point represents 1 facility within the VHA as ranked by its risk-adjusted troponin ordering rate. The risk-adjusted ordering rate is scaled to 100. Quartile 1–4 designates the facilities as separated into quartiles by their risk-adjusted troponin use.

The MOR of troponin ordering was 1.71 for the unadjusted model, 1.66 for the model adjusting for patient case mix, and 1.55 for the model adjusting for case mix and facility characteristics. Inclusion of patient case mix explained 9.5% of the variation in troponin utilization and inclusion of both case mix and facility characteristics explained 34.6% of the variation. Additional measures of patient and hospital effects of troponin ordering variability are shown in Table 3. In our sensitivity analyses, the MOR with adjustment for both case mix and facility characteristics on the whole sample set (encounters with and without laboratory orders) was 1.54. The MOR of the cohort restricted to only level 1A, 1B, and 1C facilities was 1.45.

Table 3:

Hospital effects on troponin ordering

Parameter	Model 1	Model 2	Model 3
τ2 (95% CI)	0.32 (0.24, 0.40)	0.29 (0.21, 0.36)	0.21 (0.15, 0.26)
PCV	Reference	9.5%	34.6%
ICC	0.088	0.080	0.056
MOR	1.71	1.66	1.55
C statistic (95% CI)	0.61 (0.61, 0.61)	0.71 (0.71, 0.71)	0.71 (0.71, 0.71)

CI=confidence interval. τ²=Random-effects facility-level variance; PCV=Proportional change in cluster variation calculated as $\frac{{\widehat{\tau }}_{null}^2-{\widehat{\tau }}_{full}^2}{{\widehat{\tau }}_{null}^2}$; ICC=Intracluster correlation coefficient calculated as $\frac{{\widehat{\tau }}^2}{{\widehat{\tau }}^2+\frac{{\pi }^2}{3}}$; MOR=median odds ratio calculated as $exp\left(\sqrt{2{\widehat{\tau }}^2}\times {\Phi }^{-1}(0.75)\right)$.

Model 1: Intercept only model

Model 2: Encounter level characteristics (gender, age, race, current smoking status, systolic and diastolic blood pressure, body mass index, hemoglobin, heart rate, history of the following: atrial fibrillation, alcohol abuse, anemia, coronary artery disease, congestive heart failure, chronic lung disease, cerebrovascular disease, depression, diabetes mellitus, drug abuse, hyperlipidemia, hypertension, liver disease, post-traumatic stress disorder, peripheral vascular disease, end-stage renal failure, sleep apnea, schizophrenia, bipolar disorder, ventricular arrhythmias, and history of stress test, coronary angiogram, percutaneous coronary intervention, and coronary artery bypass grafting surgery); Model 3: Model 2 + facility level covariates (Veterans Health Administration Complexity, percutaneous coronary intervention capabilities, geographic region, urban/rural status)

Association with downstream utilization

There were significant differences across quartiles in downstream testing rates with low utilization facilities trending towards lower rates of inpatient admissions (Q1:28.4% vs Q4:36.8%), echocardiograms (Q1:4.7% vs Q4:9.4%), stress testing (Q1:2.08% vs Q4:3.35%), angiograms (Q1:0.27% vs Q4:0.99%), and PCI (Q1:0.09% vs Q4:0.24%)compared to high utilization facilities (Table 4; all p<0.05). There were no statistically significant differences across quartiles for the proportion of patients undergoing CABG (Q1: 0.03% vs Q4: 0.07%; p=0.36)

Table 4:

Downstream resource use (in percentage) by quartiles of risk-adjusted troponin ordering

Parameter Median (IQR)	Q1 (N=658,087)	Q2 (N=725,483)	Q3 (N=885,379)	Q4 (N=930,375)	p-value*
Downstream utilization
Inpatient Admission	28.4 (20.5, 35.0)	36.4 (29.2, 43.1)	35.9 (30.4, 40.7)	36.8 (29.2, 42.9)	0.0021
Echocardiogram	4.71 (1.94, 7.45)	8.21 (5.73, 9.44)	8.92 (8.06, 10.09)	9.44 (7.86, 11.3)	<0.001
Stress Test	2.08 (1.28, 3.28)	2.68 (1.99, 3.53)	3.53 (2.91, 4.41)	3.35 (2.79, 4.61)	<0.001
Angiogram	0.27 (0.05, 0.87)	0.55 (0.14, 1.11)	1.03 (0.45, 1.46)	0.99 (0.23, 1.38)	0.0023
PCI	0.09 (0.02, 0.26)	0.17 (0.07, 0.47)	0.38 (0.05, 0.66)	0.24 (0.08, 0.55)	0.0367
CABG	0.03 (0.01, 0.13)	0.10 (0.02, 0.16)	0.08 (0.03, 0.15)	0.07 (0.01, 0.14)	0.3577
Falsification endpoints
KUB	2.02 (0.55, 4.11)	2.31 (0.82, 4.90)	2.66 (1.12, 4.34)	2.06 (0.85, 2.54)	0.8062
RUQ US	0.25 (0.10, 0.58)	0.30 (0.20, 0.65)	0.18 (0.03, 0.39)	0.32 (0.07, 0.48)	0.1393

Abbreviations: IQR – interquartile ratio; PCI – percutaneous coronary intervention; CABG – coronary artery bypass surgery; KUB – Kidney ureter bladder Xray; RUQ US=Right upper quadrant ultrasound

In our sensitivity analysis with re-stratification of facilities by proportion of troponin tests that resulted as critically abnormal, significant differences across quartiles was seen only for echocardiograms (p<0.01). There were no significant differences seen across facility quartiles for inpatient admissions, stress testing, angiograms, PCI, or CABG (p>0.05; Table 3 in Supplementary Materials). In our falsification endpoint analyses, there were no significant differences across facility quartiles in ordering of KUBs (p=0.81) or right upper quadrant ultrasounds (p=0.14).

Discussion

In this large nationwide study of ED visits in the VHA from 2015–2017, we found a high degree of facility-level variation in cardiac troponin testing. Patient case-mix and facility characteristics only modestly explained the variation in troponin use. Our findings suggest that higher rates of troponin utilization were associated with increased healthcare utilization.

Our study demonstrated the odds of a patient receiving a troponin test depended primarily on the hospital at which he or she received care. This finding remained consistent in our sensitivity analyses of patients with and without laboratory orders and when restricted to the highest complexity facilities. Our results add to the body of literature demonstrating high rates of variability in the use of cardiac diagnostic testing [25–27]. We suspect that in certain facilities where risk-adjusted troponin rates are high, many patients receive a defined set of laboratory orders after certain predetermined criteria are met [28]. Additionally, differences in practice culture or emphasis on patient reassurance may also play a role in driving this variation [29]. Whereas financial incentives of hospitals may normally influence testing patterns, this effect is likely minimized in VHA’s single-payor system.

We did not find a negative correlation between testing patterns of CK-MB and troponin. An observed negative correlation would suggest certain facilities preferentially use CK-MB instead of troponin. However, we did observe a modest positive correlation suggesting a cultural practice of high-volume laboratory testing. There was also no inverse correlation between troponin ordering and abnormal rates of troponin results, arguing against selectivity of troponin as a major contributor to differences in ordering rates.

Our analyses could not determine the appropriateness of an individual troponin. However, the extent of observed variation before and after risk-adjustments suggests inconsistency in troponin ordering patterns and that there may be both under and over utilization. We also observed heterogeneity within each quartile such that facilities in quartile 4 would have low rates of downstream utilization and facilities in quartile 1 with high rates of downstream utilization. Our results therefore did not assess practice patterns within individual facilities. The goal of this study was to characterize the overall national variation and potential implications of such variation and not to determine the efficiency of delivered care.

While the cost and potential harm of laboratory tests may be minimal, they can lead to further downstream testing and invasive procedures which can carry markedly higher costs and potential harm [30–32]. Our study found that facilities with higher rates of risk-adjusted troponin utilization were associated with increased rates of inpatient admissions, non-invasive testing, and invasive procedures. For some patients and their loved ones, this escalation in care may elicit intense feelings of fear, psychological distress and anxiety which may persist long after hospital discharge. Differences in rates of troponin use may be driven by a culture of frequent test ordering at certain facilities. Prior literature demonstrates that adjusted probability of biomarker test utilization is correlated with additional testing or services performed during a patient visit [33].

In our sensitivity analyses with facilities stratified by proportion of troponin tests that were designated as critically abnormal, the lack of significant variation across facility quartiles suggests that the downstream management of patients with higher probability of true disease was more homogenous and did not solely explain our finding between troponin testing and increased downstream utilization. Our falsification endpoints also did not trend in accordance with other patterns of other downstream testing, highlighting our findings are not secondary to only residual confounding.

Findings from this investigation have important implications in the context of transitioning to high-sensitivity troponin assays. Studies have shown that high-sensitivity troponin assays expedite the evaluation of patients with suspected AMI, but these studies were done in conjunction with clinical heuristics and may not be representative of practice patterns noted in the present analysis [34–35]. Recent studies demonstrate that when used without clinical assessment, high-sensitivity troponin tests have high false positivity rates which may lead to increased utilization of unnecessary downstream resources [36–37]. As our healthcare system transitions to high-sensitivity troponin testing, our study emphasizes the need to define appropriateness for troponin ordering to maximize the yield of diagnostic testing and resource utilization.

Findings from this investigation should be interpreted in the context of several limitations. First, our study is comprised of Veterans which may limit generalizability to non-Veteran populations. Second, we could not capture the characteristics of the troponin assay (e.g., Troponin I vs Troponin T). However, troponin ordering is a reflection of clinical decision making, and assay characteristics should not influence a clinician’s decision to order troponin in a patient with suspected ACS. Third, we could not track transfers of patients to outside hospitals. It is therefore possible the association of troponin to certain procedural downstream effects is an overestimate as facilities with low downstream utilization may transfer patients outside the VHA for a procedure. However, given the integrated delivery of care within the VHA the expected numbers of outside VHA facility transfers will be low. Lastly, as with all observational studies, we cannot discount the possibility that residual confounding may have affected our results. Nonetheless, we attempted to address this limitation by including falsification endpoints and sensitivity analyses.

Conclusion

We observed significant variation in troponin utilization across facilities within EDs in the VHA. This variation is only modestly explained by patient case-mix and facility characteristics. Facilities with higher troponin utilization were associated with increased downstream resource utilization. Future studies are needed to identify practices that contribute to heterogeneity in troponin utilization to guide systematic interventions to help reduce unnecessary testing and downstream resource utilization.