Outcomes Associated With High- Versus Low-Frequency Laboratory Testing Among Hospitalized Children

We performed a multicenter retrospective cohort study of children hospitalized (inpatient or observation) with 1 of 10 common conditions at 50 tertiary care referral children’s hospitals contributing data to the Pediatric Health Information System (Children’s Hospital Association, Lenexa, KS). Participating hospitals submit clinical, demographic, and billing data for all encounters, including International Classification of Diseases, 10th Revision, Clinical Modification codes. Data are deidentified at the time of submission and reviewed regularly for data quality. For this study, we excluded 4 hospitals that did not submit LOS in hours and 2 hospitals with incomplete data during the study time frame. This study was categorized as not human subjects research by the institutional review board at the lead author’s institution.

We included index hospitalizations for patients 0 to 18 years of age in 2018–2019. To focus on routine low-acuity hospitalizations for otherwise healthy children, we excluded patients with any complex chronic condition identified by using methods described by Feudtner et al,¹⁷  those with hospitalizations for newborn care, and those requiring intensive care. Furthermore, using the severity of illness indicator in All Patient Refined Diagnosis Related Groups (APR-DRGs) (3M Health Information Systems, Minneapolis, MN), we also excluded encounters with a severity of illness score >1. To capture complete hospital encounters, we excluded patients transferred to or from another acute care hospital. With this assembled cohort, we then identified encounters with the 10 most common APR-DRGs within the Pediatric Health Information System database during the study period.

We examined all complete blood cell counts (CBCs), basic chemistry studies, and inflammatory marker tests for the assembled cohort during the study period. These laboratory tests were chosen because they were the most commonly performed after a review of all laboratory tests performed for the study cohort. Laboratory studies were identified using clinical transaction codes. Individual tests and panels of tests were considered together within the 3 groupings described above (eg, an individual sodium level test, basic metabolic panel, and chemistry-7 were all considered together within the basic chemistry study category) because of differential billing practices across hospitals and to capture individual tests that may have been ordered to follow-up earlier results (Supplemental Table 3).

We defined frequency of laboratory testing as the proportion of encounter days with at least 1 laboratory test performed, that is, the total number of encounter days with at least 1 laboratory test performed divided by the total number of encounter days. We then measured overall laboratory testing frequency rates for each laboratory test grouping (CBCs, chemistry studies, and inflammatory markers) within each APR-DRG. For example, we calculated the overall testing rates of CBCs, chemistry studies, and inflammatory marker tests for all bronchiolitis encounters.

We then used the overall laboratory testing frequency rates for each laboratory test grouping and APR-DRG to compare testing frequency across hospitals. We assigned individual hospitals a score for each APR-DRG and laboratory test grouping on the basis of their relative frequency of performing laboratory tests for each condition. If the hospital’s testing rate for a laboratory test grouping and APR-DRG was <80% of the overall testing rate, it was assigned a value of −1 (ie, low rate). If the testing rate was 80% to 120% of the overall testing rate, it was assigned a value of 0 (ie, moderate rate). And if the hospital’s testing rate was >120% of the overall rate, it was assigned a value of 1 (ie, high rate). To quantify the frequency of high-rate testing, we calculated a cumulative testing score for each hospital by summing each occurrence of high-rate testing. With 3 laboratory test groupings and 10 APR-DRGs, a hospital’s maximum possible score was 30. Cumulative testing scores closest to 30 indicated the highest-frequency testing hospitals (ie, increased frequency of high-rate testing). Scores closest to zero indicated the lowest-frequency testing hospitals (ie, no or rare high-rate testing). Hospitals were divided into high-, moderate-, and low-frequency testing groups on the basis of their cumulative testing score quartile. The high- and low-frequency testing groups corresponded to the top and bottom quartiles, respectively, with the moderate-frequency testing group corresponding to the remaining 2 quartiles. Hospitals in the top quartile of testing frequency were statistical outliers (had >120% of the overall testing rate) for ≥16 of the 30 laboratory test grouping and APR-DRG combinations.

Hospitals with high-, moderate-, and low-frequency testing were compared for clinical outcomes, including LOS in hours, 7-day and 30-day emergency department (ED) revisit and hospital readmission rates, total hospital costs, and the percentage of total hospital costs attributable to the cost of each laboratory test grouping. Outcomes were adjusted for demographic and clinical factors, including age, sex, race, payer, and each hospital’s overall severity of illness. The severity of illness was assessed as a case mix index (CMI), calculated as the average relative weight for all APR-DRGs and severity of illness levels by using the Hospitalization Resource Intensity Scores for Kids score, a pediatric-specific method for measuring CMI.¹⁸ 

Categorical variables were summarized with frequencies and percentages, whereas continuous variables were summarized with geometric means and geometric SDs because of skewed distribution of data. Demographic and clinical characteristics and outcomes were compared across hospital groups by using χ² and Wilcoxon rank tests. Comparisons of outcomes between high- and low-testing hospitals were compared by using generalized linear mixed-effects models with binomial distributions for ED returns and readmission and normal distributions for cost and LOS. Cost and LOS were log transformed before modeling because of their nonnormal distribution. Models were controlled for hospital clustering through the use of random intercepts for each hospital. All statistical analyses were performed by using SAS version 9.4 (SAS Institute, Inc, Cary, NC), and P values <.05 were considered statistically significant.

We identified 132 391 encounters after application of inclusion and exclusion criteria (Fig 1). Patient demographic and clinical characteristics are summarized in Table 1. The most common age group was 0 to 4 years, almost all patients were discharged from the hospital directly home, and acuity of illness was low, as reflected in the CMI score.

The APR-DRGs included in the study are displayed in Supplemental Table 4. The included APR-DRGs composed 8.9% of hospitalizations (excluding newborn admissions) during the study years. The most common condition was bronchiolitis and respiratory syncytial virus (RSV) pneumonia; the least common condition was hypovolemia and related electrolyte disorders.

Overall testing frequency varied across groups of laboratory tests and across conditions (Supplemental Table 4). The condition with the most frequent CBC testing was cellulitis (45.0%), the one with the most frequent basic chemistry testing was gastroenteritis (66.5%), and the one with the most frequent inflammatory marker testing was cellulitis (38.3%).

Testing frequency between individual hospitals also varied across laboratory tests and conditions (Fig 2). The narrowest range of frequency of CBC testing across hospitals was for bronchiolitis (2.2%–34.0%; Supplemental Table 5); the broadest was for hypovolemia (0%–71.2%). The narrowest and broadest ranges for basic chemistry studies were in bronchiolitis (2.5%–43.1%) and hypovolemia (0%–92.6%), respectively. The narrowest and broadest and ranges for inflammatory marker testing were in bronchiolitis (0%–19.1%) and urinary tract infections (4.3%–95.3%), respectively. Infants <60 days of age accounted for 26.5% of urinary tract infection encounters.

We identified 13 hospitals with consistent high-frequency laboratory testing across conditions, 20 hospitals with moderate-frequency laboratory testing, and 11 hospitals with low-frequency laboratory testing (Fig 3). There were small differences in demographic and clinical factors in populations across low-, moderate-, and high-frequency testing hospitals (Table 1). As revealed by the heat map of testing frequency (Fig 3), hospitals tended to be consistent across conditions, for example, those with high-frequency CBC testing for one APR-DRG, also tended to have high-frequency chemistry testing for that condition. Similarly, those with high-frequency testing for one APR-DRG tended to have high-frequency testing for other APR-DRGs.

In unadjusted outcomes, the median LOS was 30.8 hours and the median hospital cost was $3050. There were small differences across low-, moderate-, and high-frequency testing hospital groups (Supplemental Table 6). Hospitals with moderate-frequency testing had the highest hospital costs and 7- and 30-day ED revisit and hospital readmission rates. Hospitals with high-frequency testing had the longest LOS. Laboratory costs accounted for a small proportion of total hospital costs, although the relative proportion of costs attributable to laboratory spending increased with higher testing frequency. After adjustment for demographic and clinical differences, there were no differences in measured outcomes comparing high- and low-frequency testing hospital groups (Table 2).

In this multicenter study of otherwise healthy children hospitalized for common low-acuity conditions, there was substantial variation in laboratory blood testing frequency across conditions and hospitals. Variation in relative testing frequency for each test group and condition was substantially lower within individual hospitals than across hospitals. Hospitals with low-frequency testing for groups of laboratory tests and individual conditions tended to also test with lower frequency for other laboratory test groups and conditions, as evidenced by the consistency of values across those factors in Fig 3. Hospitals with high- versus low-frequency testing had no differences in any adjusted clinical outcome measured, including LOS, ED revisit rate, and rehospitalization rate.

Some variation in testing across conditions can be expected. For example, the broad range of inflammatory marker testing use among patients with urinary tract infections may be partially explained by infants <60 days old making up more than one-quarter of encounters for this condition and the known variation in inflammatory maker testing use in evaluating febrile infants.¹⁹  As another example, the higher rate of electrolyte testing in gastroenteritis may be partially explained by guideline recommendations for laboratory testing in areas of diagnostic uncertainty,²⁰  compared to bronchiolitis, for which guidelines generally recommend no testing.²¹  But given the low acuity of our patient cohort and the gastroenteritis guideline recommendation that “supplementary laboratory studies, including serum electrolytes, to assess patients with acute diarrhea usually are unnecessary,”²⁰  the 67% overall chemistry testing rate in gastroenteritis in our study almost certainly represents significant overtesting. Similarly, there is little evidence to support the 45% overall frequency of CBCs in low-acuity cases of cellulitis and urinary tract infection or the 38% frequency of inflammatory marker testing in cellulitis cases. Additionally, the broad range in frequency of laboratory testing across hospitals for specific conditions (eg, the almost 0% to 100% range in C-reactive protein testing for urinary tract infection) suggests overtesting at some institutions. These specific results suggest some potential key targets for future quality improvement efforts.

The consistency of relative testing frequency within hospitals across conditions and laboratory test groupings suggests a strong role for institutional culture in guiding testing decisions. High variation in testing frequency, despite lack of evidence for test utility, has been demonstrated in specific pediatric conditions^13–15  and, for electrolyte testing, in a general pediatric population.¹⁶  A number of factors may play a role in determining institutional testing patterns, including hospital volume,²²  cost transparency,²³  place of faculty residency training,²⁴  and presence or absence of faculty champions of high-value care.²⁵  Both the high overall testing rates for common laboratory tests and conditions and the institutional variation in testing rates suggest significant overuse of laboratory tests.

We hypothesized that hospitals with high-frequency testing patterns would be associated with worse clinical outcomes, including longer LOS and higher costs, for common conditions. Some previous studies of specific conditions did find associations between higher testing frequency and clinical outcomes. For example, Florin et al¹³  noted higher odds of hospitalization for hospitals with high use of diagnostic testing of all types (including laboratory and radiology studies) in children with community-acquired pneumonia evaluated in the ED. Similarly, Lind et al¹⁵  identified higher risk of hospitalization among hospitals with higher-frequency testing patterns in children treated in the ED for gastroenteritis. Thomson et al¹⁴  reported a higher readmission rate associated with hospitals with higher testing frequency in children with neurologic impairment and pneumonia. It should be noted, however, that of the 2 studies reporting LOSas an outcome, neither found an association with higher testing frequency.^14,15  In our study population, which was characterized by low patient acuity and a short (31-hour mean) LOS, it is perhaps unsurprising that we would find no differences in clinical outcomes. And although we identified higher laboratory costs in hospitals with high-frequency testing, the lack of difference in overall cost likely reflects the small proportion of hospital costs attributable to laboratory tests. In any case, we identified no evidence that high-frequency testing in this population was associated with any improvement in outcomes. Hospitals with lower-frequency testing achieved the same outcomes, which suggests that high-frequency testing represents overuse and argues strongly for more judicious use of laboratory testing for children with the common pediatric conditions we studied.

Our study should be interpreted in light of several important limitations. Data were derived from an administrative database, which limits some important clinical context that might have influenced decision-making for laboratory testing use in specific cases. The data set only includes children’s hospitals, which may differ in important respects from other settings where children receive health care. The retrospective study design precludes our ability to prove any associations we identified as causally related. In particular, it is possible that unmeasured factors, including differences in clinical severity or other patient factors across the hospitals studied, contributed to our findings, although it unlikely that unmeasured differences alone could account for the large ranges in testing frequency we identified across hospitals, and our inclusion of only patients with low severity of illness scores should mitigate or preclude any large unmeasured differences in clinical characteristics. Lastly, we are unable to assess encounters for which increased laboratory testing may be appropriate, such as in patients on multiple nephrotoxic medications.

In a multicenter study of otherwise healthy children hospitalized for common low-acuity conditions, we identified substantial variation in frequency of use of common laboratory tests. There was more variation across hospitals than within hospitals for particular conditions and tests, suggesting a strong role for institutional culture in determining testing patterns. Important clinical outcomes, including LOS and readmission rates, were not different between hospitals with high- and low-frequency testing patterns. Our results suggest opportunities to reduce overuse of common laboratory tests among hospitalized children and highlight the need for broad quality improvement efforts to improve high-value care delivery in this population.