Correlating Scores but Contrasting Outcomes for Eat Sleep Console Versus Modified Finnegan

In this retrospective cohort study, we identified infant-mother dyads at risk for NOWS between 2016 and 2019 across a hospital system encompassing 1 academic and 2 community level I and II nursery settings. In this study, we compared simultaneously obtained M-FNASS and ESC scores as well as health care use outcomes when infants were managed with an M-FNASS–based versus ESC-based clinical algorithm. This study was approved by the institutional review board of the affiliated medical college, which determined that informed consent was not necessary for this retrospective review.

Infants were admitted for their birth hospitalization to 1 of 3 newborn nursery sites in our hospital system. Site 1 was an academic well-newborn nursery with a level II special care nursery and was located within a tertiary care and high-risk obstetrics and gynecology referral center. This location cared for ∼3000 births per year during the study period. Site 2 was a community level I well-newborn nursery with an average of 670 births per year. Site 3 was a community well-newborn nursery with a level II special care nursery that cared for an average of 810 births per year. Maternal substance use during pregnancy did not necessitate transfer to the high-risk obstetrics and gynecology service at our academic center, but traditionally, our academic center cared for mother-infant dyads affected by NOWS more frequently than the 2 community sites. General pediatricians, family practice physicians, neonatologists, neonatal advanced practice providers, obstetricians and gynecologists, nurses, lactation consultants and social workers collaborated in the care of infant-mother dyads at all 3 sites.

Recommended management of an infant at risk for NOWS was consistent across all 3 nursery locations within our hospital system. For infants with confirmed or suspected prenatal exposure to opioids, providers were encouraged to obtain urine toxicology testing on the mother during labor as well as urine and meconium toxicology testing on the infant after birth. Providers clinically monitored the infant inpatient for signs of NOWS for a minimum of 5 days. Infants roomed-in with their mother unless their clinical acuity required admission to the level II special care nursery or other extenuating circumstances arose. Nurses assessed the infant every 3 to 4 hours for symptoms of NOWS and documented scores in the electronic medical record (EMR).

During epoch 1 (January 2016 to July 2018), infants were exclusively scored and managed by using an M-FNASS–based algorithm (Supplemental Fig 3). During epoch 2 (August 2018 to September 2019), infants were simultaneously scored with the M-FNASS and ESC approach and managed by using an ESC approach (Supplemental Fig 4). Copyright permission was obtained to document ESC scores in the EMR during epoch 2. The transition from M-FNASS–based management to ESC-based management was made across all 3 included nurseries at the same time, preceded by a multidisciplinary education campaign involving nurses, advanced practice providers, and physicians caring for infant-mother dyads at each site.

When indicated, weight-based dosing of oral morphine was used for pharmacologic management of withdrawal symptoms. Patients were eligible for pharmacologic therapy during epoch 1 if they demonstrated ≥3 consecutive M-FNASS scores ≥8 or ≥2 scores ≥12. Infants were eligible in epoch 2 if they demonstrated ≥2 consecutive ESC scores containing at least 1 “no” component. Dose escalation and deescalation were standardized and consistent across both epochs (Supplemental Figs 3 and 4).

For data collection, we used the I2B2 Cohort Discovery Tool (Partners HealthCare System, Boston, MA) to identify eligible dyads by searching for the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) and International Classification of Diseases, 10th Revision, Clinical Modification (ICD-10-CM) codes (779.5 and P96.1 [“neonatal abstinence syndrome”], 760.79 and P04.14 [“newborn affected by maternal use of opiate”], and P04.9 and 760.7 [“intrauterine drug exposure”], respectively) documented for inpatient birth hospitalization encounters between January 1, 2016, and September 30, 2019, querying both problem list and billing diagnoses. Infants were excluded if there was no prenatal opioid exposure identified, if NOWS scores were not documented, or if the infant was transferred to another hospital system. Patient chart data extraction was performed by using Honest Broker (Clinical and Translational Science Institute of Southeast Wisconsin Bioinformatics Team, National Institutes of Health grant 5UL1TR001436-02). Five reviewers (K.R., A.M., M.G., K.S., and E.C.) then mined each mother’s and infant’s EMR charts to complete demographic, NOWS scoring, and health care use data collection.

In the statistical analysis, we compared M-FNASS and ESC scores and outcomes between epochs 1 and 2 using SAS (version 9.4; SAS Institute, Inc, Cary, NC). The relationship between ESC and M-FNASS scoring components was characterized by using generalized linear mixed modeling to account for repeated measurements from the same subjects. Specifically, the probability of escalating or deescalating M-FNASS scores was modeled with the ESC score as the predictor by using ordered multinomial logistic regression. The Holm-Bonferroni correction was then performed to control for type 1 error in multiple comparisons. P values <.05 were considered significant. A receiver operating characteristic (ROC) curve was calculated for all scores by using a generalized estimating equation with logit link function and binomial distribution to model the probability of M-FNASS scores corresponding to an ESC score containing at least 1 “no” component. Subjects were used as clusters with the first-order autoregressive covariance structure for the repeated-measures design. An ESC score containing at least 1 “no” component was deemed a threshold of particular clinical significance because an infant with consecutive scores containing ≥1 “no” would be eligible for initiation of pharmacotherapy.

Health care use outcomes from the 2 epochs were compared by using the χ² test or Fisher’s exact test for categorical variables and the Mann–Whitney U test for continuous variables. Comparisons of length of stay and initiation and duration of pharmacologic treatment were adjusted for significant covariates by using multivariable generalized linear models. γ distribution with log link function was used for length of stay and duration of pharmacologic treatment analyses because of the skewed nature of the data (recommended minimum 5-day length of stay for infants at risk for NOWS). Binomial distribution with logit link function was used for the analysis of initiation of pharmacologic treatment.

A total of 180 infant-mother dyad charts were identified for review. Ten infants were excluded because of the lack of any documented prenatal opioid exposure. Ten infants were excluded because of a transfer outside the hospital system. Two infants were excluded because no withdrawal scoring was documented (Fig 1). A total of 2101 simultaneous M-FNASS and ESC scoring instances were included for analysis.

Demographics of included infant-mother dyads were largely similar between epochs, except for site of hospitalization, maternal referral for substance use treatment, and infant meconium toxicology test result (Table 1). Although most infant-mother dyads were hospitalized at our academic site (site 1) over the course of this study, more infants were hospitalized at community sites (sites 2 and 3) during epoch 2 when compared with epoch 1. During epoch 2, more mothers were referred for ongoing inpatient or outpatient management of substance use after their delivery hospitalization when compared with epoch 1. More infants had a positive meconium toxicology test result (for any opioid or nonopioid substance) in epoch 2 when compared with epoch 1, although there was no significant difference in the number of positive urine toxicology test results.

Table 2 outlines the strength of association between domains of the ESC scoring system and the M-FNASS. Each value is the probability that the 2 scoring domains are not correlated; thus, a smaller value in the table signifies a stronger association between the 2 scoring systems. Each ESC domain (eat, sleep, and console) significantly correlated with the overall M-FNASS score. ESC scores containing all “yes” components were significantly associated with lower total M-FNASS scores, in contrast to ESC scores with at least 1 “no” component, both in overall scores and in 11 of 21 individual M-FNASS domains. The cry, sleep, and Moro reflex domains, as well as the total M-FNASS score, consistently correlated with all ESC scoring components. Ten M-FNASS domains (disturbed tremor, undisturbed tremor, tone, stuffiness, sneezing, respiratory rate, sucking, feeding, vomiting, and stools) significantly correlated with some components of the ESC model but not all. M-FNASS domains for sweating, fever, yawning, and mottling did not correlate with ESC domains. Of note, a “yes” to the eat domain of the ESC model correlated with a higher M-FNASS score for excoriation. Correlations were not able to be assessed for myoclonic movements, convulsions, or nasal flaring because these were rare (or never) occurrences in our data set, providing insufficient variability for analysis.

A ROC analysis revealed an optimal cutoff of 7.5 (sensitivity = 0.84, specificity = 0.70, area under the curve = 0.842) when modeling the probability of M-FNASS scores corresponding to an ESC score containing at least 1 “no” component (Fig 2).

Infants who were managed with the ESC approach compared with the M-FNASS approach demonstrated reduced length of stay (median 9.5 vs 5 days; P = .0002) and reduced use of pharmacologic treatment (53% vs. 33%; P = .018). For those infants who were started on pharmacologic treatment, duration of pharmacologic treatment (median 11 vs 7 days; P = .0042) and length of stay (median 15 vs 10 days; P = .0002) were also significantly reduced (Table 3).

Adverse events were rare and did not increase from epoch 1 to epoch 2. Four infants (3 in epoch 1 and 1 in epoch 2) required transfer to a higher level of care for neurologic indication. Five infants (4 in epoch 1 and 1 in epoch 2) who were receiving oral morphine experienced brief self-resolved bradycardic events. One infant’s family refused morphine administration during epoch 2, and the infant was monitored for 5 days and discharged from the hospital because withdrawal symptoms self-resolved. A retrospective chart review did not reveal any infants readmitted because of NOWS within 30 days of discharge from their birth hospitalization.

In this study, we examine how M-FNASS and ESC scores relate to one another in clinical practice, allowing providers to consider what is and is not equivalent between the 2 systems. Although granular, this comparison between scoring systems provides the crucial context with which to understand a growing body of literature revealing reduced health care use with an ESC approach to NOWS.^{23–26,29–38}  NOWS management remains highly varied across the United States.^39–43  Within our own hospital system, we encountered hesitancy from a variety of members of our multidisciplinary health care team regarding transition from the well-known M-FNASS model to the newer ESC approach. Without evidence that the ESC approach similarly identifies and manages infants who are symptomatic, providers may hesitate to change their clinical practice even with published improved health care use outcomes.

Our findings reveal that ESC scores consistently correlate with overall as well as individual M-FNASS domains. In particular, ESC scores correlated with components of the M-FNASS that have previously been found to predict NOWS severity (eg, tremor, tone, Moro reflex, and sleep) and did not correlate with components of the M-FNASS that have not been found to predict NOWS severity (eg, fever, mottling, sweating).^{14–16,22,44} 

Commonly used thresholds for initiating pharmacotherapy for NOWS systems also correspond closely between M-FNASS and ESC scoring systems. Serial M-FNASS scores of ≥8 and serial ESC scores containing at least 1 “no” are both used clinically as evidence of significant disease severity and as criteria for initiating or escalating pharmacologic therapy for NOWS. By using a ROC analysis, we have demonstrated that an ESC score containing at least 1 “no” clinically corresponds to an M-FNASS score of ∼7.5. Thus, this clinical threshold of disease severity is similarly detected by both scoring systems. Any difference in pharmacotherapy use between systems, therefore, does not appear to be due to an inherent difference in detection of clinical disease. Management algorithms dependent on trending M-FNASS scores could thus transition to an ESC model without sacrificing detection of escalating disease severity.

A transition from the M-FNASS to the ESC approach for infants with NOWS resulted in reduced inpatient length of stay and reduced dependency on pharmacologic management strategies, without detected increased adverse events in our cohort, reaffirming previously published results by other authors at other institutions. Our data also reveal that this reduction in length of stay occurs both in infants who receive pharmacologic treatment of NOWS and those who do not.

In summary, the correspondence between overall ESC and M-FNASS scores and pharmacotherapy initiation thresholds presented here suggests that the 3-part ESC system is sufficient in comparison with the 21-part M-FNASS for clinical monitoring and management of NOWS. These corresponding scoring systems, however, produce divergent clinical outcomes in our and others’ cohorts. This suggests that the reduced health care use outcomes associated with the ESC approach are not the result of underdetection or undertreatment of NOWS. Rather, the ESC paradigm may facilitate higher-value inpatient care.

Our study has several limitations. Fundamentally, the M-FNASS and the ESC approach are both subjective assessments of a clinical syndrome. We felt that comparing M-FNASS and ESC scores obtained at the same time by the same nurse on the same infant would best mitigate the influence of this subjectivity on our data, similar to the approach undertaken by Grossman et al²⁶  in their 2018 published comparison of FNASS versus ESC scoring outcomes. Transitioning to the ESC model while continuing to document M-FNASS scores for a limited amount of time, in our experience, also mitigated some provider anxiety regarding the change in NOWS management. This meant, however, that both M-FNASS and ESC scores were visible to the health care team during epoch 2. Although providers were instructed to only use ESC scores to make clinical management decisions, it is possible that viewing the M-FNASS scores may have influenced treatment decisions. If present, this influence would have likely made health care use outcomes between epochs 1 and 2 more similar. Thus, the reduction in health care use associated with an ESC approach in comparison with the M-FNASS may be underestimated in our analysis.

Additionally, retrospective cohort studies relying on chart review for data collection are inherently susceptible to limitations in data accuracy and availability, as evidenced by small numbers of missing data in our demographic analysis. Identifying patients affected by NOWS by billing code has been demonstrated to be a reasonable approach but may also underestimate the disease burden within a population.⁴⁵ 

We also did identify statistically significant differences between the infant-mother dyads managed in epoch 1 versus epoch 2. We hypothesize that increased numbers of infant-mother dyads cared for in community hospital settings and increased maternal referral to substance use treatment after discharge represent improvements in the patient-centered care of these dyads. We speculate that these effects may be related to implementation of a new standardized protocol (and preceding multidisciplinary educational efforts) rather than inherent differences between populations. The increased rate of positive meconium toxicology test results in epoch 2 raises concern that infants in epoch 2 may have had more prenatal substance exposure, thereby putting them at risk for more severe NOWS. However, this trend is not corroborated by other predictors of NOWS severity, such as positive urine toxicology test results or rapid newborn weight loss.

A comparative analysis could not be used characterize the relationship between nasal flaring, convulsions, and myoclonic movements with the ESC model because of low variability. These symptoms are generally considered severe manifestations of uncontrolled withdrawal. We speculate that their low variability within our epoch 2 cohort reflects effective symptom management with the ESC approach.

Lastly, we found adverse events to be rare and readmission for NOWS to be nonexistent in our study population, before and after intervention, and thus we cannot attribute these low rates to our intervention. However, we hope our experience is reassuring for practitioners considering a transition to the ESC model of NOWS care.

By directly comparing M-FNASS and ESC scores and subsequent health care use metrics, in this study, we demonstrate that the ESC model meaningfully correlates with the M-FNASS as a NOWS scoring system, similarly signals when NOWS symptoms may warrant pharmacologic management, and is associated with reduced health care use for infants at risk for NOWS in the nursery setting. This suggests that the ESC model facilitates similar monitoring but more effective management of NOWS.