Trends and Variation in Length of Stay Among Hospitalized Febrile Infants ≤60 Days Old

Data for this multicenter retrospective cohort study were obtained from the Pediatric Health Information Systems (PHIS), a multicenter administrative database. PHIS contains demographic, diagnosis and billing data, including International Classification of Diseases Codes, 10th revision, Clinical Modification (ICD-10-CM), from the 49 hospitals affiliated with the Children’s Hospital Association (Lenexa, KS). Data are deidentified, and unique encrypted identifiers enable tracking of individual patients across visits within sites. The Children’s Hospital Association and participating centers jointly assure data quality and integrity.¹⁸  For this study, we excluded hospitals that did not report LOS in hours (n = 7) and those with incomplete data during the study period (n = 3). The study was deemed exempt from review according to policies at the lead author’s institutional review board.

We included infants aged ≤60 days hospitalized (inpatient or observation status) between January 1, 2016, and December 31, 2019, with ICD-10-CM codes P81.9 (disturbance of temperature regulation of newborn), R50.81 (fever presenting with conditions classified elsewhere) or R50.9 (fever, unspecified) as a primary or secondary diagnosis. The billing codes for fever were mapped from previous studies by using International Classification of Diseases Codes, Ninth Revision, Clinical Modification codes.^18,19  The time frame was chosen to include continuous data after transition from the International Classification of Diseases Codes, Ninth Revision, Clinical Modification to the ICD-10-CM classification system. We excluded initial newborn hospitalizations, infants with complex chronic conditions determined by the methodology developed by Feudtner et al,²⁰  and infants with an ICU admission or who were transferred to or from other hospitals (including infants not discharged from the hospital). Additionally, to focus on infants with undifferentiated fever, we excluded patients diagnosed with specific infections that would be associated with longer LOS (Supplemental Table 2), adapted from methods of Nguyen et al¹⁷  and Aronson et al,¹⁹  modified for the ICD-10-CM system. Thus, as examples, we excluded infants with an ICD-10-CM diagnosis code for bacterial infections or for viral infections associated with lower respiratory tract disease, such as bronchiolitis or viral pneumonia. We included other viral conditions, including respiratory syncytial virus or influenza infections not involving the lower respiratory tract, given that most febrile infants are eventually determined to have viral infections.

The primary outcome was the trend in LOS. To compare the trend in LOS for hospitalized febrile infants with overall trends in hospital LOS, we also assessed the LOS for all other medical (ie, nonsurgical, excluding the study cohort) hospitalizations during the study time frame for patients aged ≤60 days, with the same exclusions as the study cohort (eg, ICU care, complex chronic condition). We also examined trends in LOS for individual hospitals, adjusted for patient demographics, including age, sex, race, ethnicity and payer, along with each hospital’s overall severity of illness. Severity of illness was assessed as a case mix index (CMI), calculated as the average relative weight for all All Patient Refined Diagnosis Related Groups and severity of illness levels using the Hospitalization Resource Intensity Scores for Kids (HRISK) score, a pediatric-specific method for measuring CMI.²¹  Lastly, we assessed for any relationship between LOS and 30-day all-cause readmission rate for the study cohort at each hospital, also adjusted for patient demographics and severity of illness.

Demographic characteristics were summarized by using frequencies and percentages, and the geometric mean LOS was compared across strata of characteristics by using Wilcoxon rank tests. We used geometric means because of the skewed nature of LOS and log-transformed LOS before modeling. We assessed for trends in LOS using generalized linear mixed effects models with random intercepts for each hospital and fixed effects for time (year or quarter), patient-level demographic (age, race, ethnicity, and payer) and clinical characteristics (HRISK). The comparison of the trend in LOS for febrile infants versus other medical hospitalizations was assessed by using an interaction term between time and an indicator for cohort (febrile infant versus other medical). We additionally modeled each hospital’s LOS trend (without the random effect) and estimated and compared their adjusted LOS in 2016 vs 2019, overall and stratified by age (0–30 days and 31–60 days). All statistical analyses were performed by using SAS v. 9.4 (SAS Institute, Inc, Cary, NC), and P < .05 was considered statistically significant.

We identified 11 868 eligible encounters for febrile infants aged ≤60 days after study exclusions (Fig 1). Bronchiolitis and “unspecified Escherichia coli as cause of diseases classified elsewhere” accounted for 48.5% of excluded infections. The characteristics of included study patients are summarized in Table 1. More than half of the encounters (60%) were for infants aged ≤30 days. The unadjusted mean LOS for the study cohort was 42.2 hours. Infants aged ≤30 days had longer LOS compared with those aged 31 to 60 days (45.5 vs 37.7 hours, respectively [P < .001]).

The adjusted LOS for the study cohort decreased from 44.0 hours in 2016 to 41.9 hours in 2019 (P < .001) (Fig 2). The trends in LOS for infants ≤30 days and 31 to 60 days old were not different (P = .37, Supplemental Fig 4). There was also a significant trend downward in LOS for all other medical hospitalizations during the study period (P < .001), and the 2 trends were not significantly different (P = .35) (Fig 2).

The adjusted geometric mean LOS varied substantially across hospitals, ranging from 33.5 to 77.9 hours in 2016 and from 30.4 to 100.0 hours in 2019 (Supplemental Table 3) (Fig 3). The change in adjusted mean LOS between 2016 and 2019 across individual hospitals also varied widely, from −23.9 to +26.7 hours, with a median change of −1.8 (interquartile range: −5.4 to 0.3). Ten hospitals had a statistically significant change in LOS between 2016 and 2019; 8 hospitals decreased LOS and 2 increased LOS (Fig 3). Most hospitals had a shorter LOS at the end compared with the beginning of the study period (Fig 3).

The median 30-day readmission rate for the study cohort was 4.2% (interquartile range: 1.8 to 6.0) across hospitals and ranged from 0% to 19.2%. There was no relationship between adjusted LOS and adjusted 30-day readmission rate (P = .70).

In this multicenter study, we identified an ∼2.1-hour shorter LOS for hospitalized febrile infants aged ≤60 days in 2019 compared with 2016. We also identified substantial variation in LOS for febrile infants across hospitals within a given year (eg, a 69-hour difference in LOS among participating hospitals in 2019). The change from 2016 to 2019 in LOS in individual hospitals also varied substantially, with 20% of included study hospitals having a statistically significant decrease in LOS. There was no association between shorter LOS and hospital readmission rate.

The overall decrease in febrile infant LOS has a number of possible explanations. As we noted in the introduction, a multicenter collaborative, REVISE, in which several of our study hospitals participated during the study period, was implemented to reduce variation in care of febrile infant and set 42 hours as a target LOS from initial vital signs to discharge.¹⁶  However, participation in REVISE does not fully explain the findings we observed, given that only 18 of the 39 institutions in our study participated in the collaborative during the study. It is also possible that practitioner awareness of the collaborative, or recognition of the emerging data suggesting safety of shorter LOS in the febrile infant population, affected LOS at other institutions. The downtrend in LOS that we observed may also be a product of overall downward trends in LOS in hospitalized children, which have been reported for a number of conditions, including skin and soft tissue infections²²  and pneumonia.²³  However, the 2.1-hour decrease in LOS we identified, although statistically significant and perhaps a good start for improvement, is of limited clinical importance for most individual patients and their caregivers. Given the abundance of emerging data supporting the safety of a shorter LOS for febrile infants and the harms associated with unnecessary hospitalization,^24,25  we expected a greater decrease than observed, suggesting the need to more rapidly scale and sustain quality improvement efforts.

The wide variation across individual children’s hospitals also suggests a strong role for institutional culture and practice patterns in determining LOS. Similar variation in practice has been observed across children’s hospitals in a number of different conditions, including measurement of serum electrolytes,^26,27  testing with inflammatory markers for infections,²⁸  and evaluation and treatment of children with pneumonia^29,30  and bronchiolitis.³¹  The variation in febrile infant LOS across institutions, with no increase in readmissions associated with shorter LOS, suggests significant opportunity for institution-specific quality-improvement efforts.

The only other recent study comparing trends in febrile infant LOS did not identify decreases over time.¹⁷  As we noted in the introduction, this previous study differs from ours in several important respects, including the study time frame ending in 2012, measurement of LOS in days, and use of the Kids Inpatient Database, which contains a broader range of institutions, including community hospitals, than the PHIS database. The REVISE collaborative, which employed a number of quality initiatives to standardize the care of febrile infants aged 7 to 60 days across 124 hospitals (community and academic) in 34 states, did reveal a postimplementation reduction in LOS similar to that observed in our study, with a reduction of 5.3% overall, decreasing from a median of 50 to 46 hours.¹⁶ 

Our study should be interpreted in light of several limitations. The retrospective design precludes our ability to account for all factors that could potentially influence LOS. Although we accounted for demographic characteristics and hospital severity factors, there may be unidentified confounders that separately influenced LOS at individual institutions. Billing codes have good accuracy in identifying febrile infants,¹⁹  but this methodology has imperfect sensitivity and specificity and likely misclassified some patients. Additionally, important clinical information that could affect LOS, including time of culture collection, is not available in the PHIS database. The time frame included in our study, 2016–2019, although having the advantage of having consistent billing codes to identify patients, may be shorter than ideal to capture all recent trends in LOS, and, in particular, may have missed changes between 2012 and 2015, after publication of important articles suggesting safety of shorter LOS.^7,8,15  We only examined all-cause readmissions and, thus, did not assess for any variation in readmissions due to planned rehospitalizations. Lastly, the use of the PHIS database containing only freestanding children’s hospitals may limit generalizability of our findings to other settings.

In our multicenter study of children’s hospitals, we identified a small downward trend in LOS among febrile infants aged ≤60 days hospitalized between 2016 and 2019. We also identified substantial variation in febrile infant LOS across individual children’s hospitals, with no association between LOS and hospital readmissions. Our results, although encouraging, suggest a need for continued broad quality-improvement efforts to reduce variation in institutional LOS among hospitalized febrile infants.