Early Postnatal Discharge for Infants: A Meta-analysis

The full systematic review protocol has been published⁵  and registered in PROSPERO (registration CRD42015020545). This review conforms to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses statement.

Electronic databases (CENTRAL, Medline, Embase, Cumulative Index to Nursing and Allied Health Literature, and SCI) were searched for articles up to January 15, 2018, with the following search terms: postnatal care, postpartum period, puerperium, postpartum, length of stay, patient discharge, hospital stay*, and patient readmission. In the searches, we used free text and indexed terms combined by using Boolean operators, adjusted for each database (Supplemental Table 3). The search strategy was not limited by study design and time, language, or geographical restrictions. When applicable, authors of primary studies were contacted for further information.

All studies were eligible to be included in the review if they met the following criteria: women and infants considered “fit for discharge” by their health care practitioners; all births occurring in either obstetric-led (for which obstetricians have primary professional responsibility for women at high risk of complications during labor and birth and women are cared by a team of midwives and doctors) or midwife-led care settings (in which midwives take the primary professional responsibility for the labor care for low-risk women); studies had to compare a policy of early discharge from hospital in which early discharge referred to a hospital discharge that was <48 hours after birth (or <96 hours for cesarean delivery) and earlier than the standard care in the setting in which the intervention is implemented to be included. In the Cochrane review,³  the authors identified that RCTs alone on this topic are limited by poor compliance and clinical and methodologic heterogeneity, indicating that alternative study designs may be more helpful for assessing the effect of policy change. Therefore, as guided by Cochrane Effective Practice and Organisation of Care (EPOC),⁴  in addition to RCTs, non-RCTs, controlled before-after studies (CBAs), and interrupted time series (ITSs) were also eligible for inclusion in the review. ITSs can provide a method of measuring the effect of an intervention when randomization or identification of a control group are impractical.⁴  Multiple data points are collected before and after the intervention, and the intervention effect is measured against the preintervention trend.⁴ 

Infant and maternal outcome measures were guided by the Cochrane review.³  The primary infant outcomes were the proportion of infants readmitted to the hospital within 7 days and within 28 days after birth. Maternal outcomes were the proportion of women readmitted for complications related to childbirth (postpartum hemorrhage, retained products of conception, infection, postpartum psychosis); the proportion of women breastfeeding (exclusively or partially) at 48 hours, 6 weeks, and 6 months after birth; and the proportion of women with a score indicating probable depression on a validated standardized instrument for measuring depression.

Citations were screened for inclusion independently and in duplicate. Articles were assessed unblinded, and differences in opinion were resolved through discussion or a third researcher. Data extraction was performed independently and in duplicate. The EPOC data collection form⁶  was adapted to answer review specific research questions. Data extraction forms were piloted on a sample of included studies. The methodologic quality of included studies was assessed independently and in duplicate by using EPOC criteria for the risk of bias tool⁷  and Cochrane RCT tool as appropriate.⁸ 

Meta-analyses of RCT studies were conducted in RevMan (version 5.3)⁹  by using a random effects model, and, when significant heterogeneity was present, data were described in a narrative synthesis. When data from ITSs were presented graphically, data were extracted from graphs by using Plot Digitizer software¹⁰  and reanalyzed by using autoregressive integrated moving average with SPSS version 22 (IBM SPSS Statistics, IBM Corporation)¹¹  as described in EPOC guidance.¹² 

We used the autoregressive integrated moving average analysis to estimate the effect of a policy change while taking into account the time trend and autocorrelation among the observations. Estimates for the regression coefficients correspond to 3 standardized effect sizes: change in level at 1 and 2 years post policy change and change in slope. The change in level was defined as the difference between the observed level at the intervention time point and that predicted by using the preintervention time trend.¹²  The change in slope was defined as the change in trend from pre- to postintervention reflecting the long-term effect of the policy intervention.¹²  Data were then standardized by dividing the level and slope and SE by the SD of the preintervention slope. The effect sizes for change in the level at 1 and 2 years and effect size for change in slope were entered into RevMan 5 and meta-analyzed by using the generic inverse variance method with random effects. The statistical significance was set at P <. 05. When data from the same region and time period were used in different studies, only 1 study was used for inclusion in the meta-analysis.

For RCTs, statistical heterogeneity was examined by inspection of confidence intervals (CIs) and the I² statistic. Heterogeneity was also explored through subgroup analysis. Sensitivity analyses were planned to assess the effect of incomplete outcome data and fixed effects versus random effects analysis. On the basis of the Cochrane review, subgroup analyses were undertaken to compare trials with cointerventions (such as increased home visiting in the early discharge group or no cointervention). Further subgroup analyses were planned to explore the effect of the mode of birth, timing of discharge, type of hospital delivered at (consultant-led unit or midwife-led unit), gestation at birth, and level of risk (high risk or low risk); however, these were not possible because of lack of participant level data and small number of participants in most trials. For the ITSs, study methods to adjust for potential confounders were taken into account in interpreting results. There were insufficient studies to assess publication bias through the use of funnel plots.¹³ 

We found 9303 studies from electronic sources and hand searches. After removal of duplicates and eligibility screening, 15 studies were identified for inclusion (Fig 1). Ten were RCTs that took place between 1976 and 2015 in several countries, including Canada, England, Egypt, Malaysia, Spain, Sweden, Switzerland, and the United States. All studies were conducted in obstetric-led hospital settings.^14–23  In the trials, the effects of a policy of early postnatal discharge were compared to a standard length of postnatal stay for women and infants. In contrast, in 4 population-based cohort studies with ITS analyses, researchers assessed the effect of state and federal legislation introduced in the United States prohibiting insurance plans from limiting coverage for postpartum hospital stay to <48 hours for normal vaginal deliveries and <96 hours for cesarean sections on various health related outcomes.^24–27  Before the legislation introduced in 1996–1997, there was no minimum length of postnatal stay in the United States. In the fifth ITS study, researchers examined the effect of a same-day discharge policy in 5 Danish counties introduced over the period 1990–2003.²⁸  All study characteristics are summarized in Table 1, and Supplemental Table 7 illustrates which studies reported on the outcomes described in the review. Maternal outcomes are described in Supplemental Figs 4–8.

The pooled result of the 7 trials that reported on infant readmission to hospital within 28 days after birth revealed that infants were significantly more likely to be readmitted to hospital within 28 days after birth if they were discharged from the hospital <48 hours compared with infants discharged >48 hours (risk ratio [RR]: 1.70 [95% CI 1.34 to 2.15]; Fig 2).^{14–18,20,23}  A planned subgroup analysis of RCTs with a cointervention and RCTs with no cointervention was conducted. In the 2 trials without a cointervention, the researchers happened to include only women who had cesarean delivery. These subgroups revealed no change in the direction of the effect: the no cointervention RR was 1.61 (95% CI 1.00 to 2.68; I² = 14%) compared with the cointervention RR of 1.74 (95% CI 0.82 to 6.68; I² = 0%). There were insufficient data to provide a meta-analysis for readmissions within 7 days. This outcome was only reported in 1 study, in which the authors reported a RR of 3.24 (95% CI 0.13 to 77.63) in favor of the control group.¹⁸ 

In results from 4 primary ITS studies in the United States, in contrast to the RCTs, researchers looked at the reverse intervention: a policy of a minimum postnatal length of hospital stay. Datar and Sood²⁵  found that once a minimum legislation was introduced, there was a significant reduction in the odds of neonatal readmission in California from −9.3 per 1000 live births in the first year postlegislation to −11.8 per 1000 live births in second year postlegislation and −19.7 per 1000 live births in the third year postlegislation [P < .01]). This trend was observed across all subgroups including mother’s education, mother’s age at birth, race, parity, delivery type, and antenatal complications (Supplemental Table 4). Evans et al²⁴  found that in California the legislation was most beneficial for infants of cesarean delivery, complicated vaginal delivery, or a Medicaid recipient with complicated vaginal delivery, with little evidence that readmission rates were reduced for newborns from uncomplicated vaginal deliveries. Madden et al²⁹  and Meara et al²⁶  found no significant change over the prelegislative, legislative, or postlegislative period for neonatal readmission to hospital. In the fifth ITS study, which was conducted in Denmark, Sievertsen and Wüst²⁸  found that the same-day discharge policy resulted in a 3% increase in infant readmission rates within 28 days of birth (0.031; SE: 0.11; P < .01; Supplemental Table 4).

The results of the meta-analysis of 3 digitized and then reanalyzed ITSs reveal that when the preslope trend was taken into account, there was a decrease in the proportion of infants readmitted within 28 days after the minimum postnatal stay policies and legislation were introduced, and this became statistically significant in the first and second year (change in slope: −0.62 [95% CI −1.83 to 0.60]; change in level first year: −4.27 [95% CI −7.91 to −0.63]; change in level second year: −6.23 [95% CI −10.15 to −2.32]; Fig 3).

The I² statistic for assessment of heterogeneity was 19% for the change in level and 0% for the change in level at 1 year and change in level at 2 years: therefore, further investigations for heterogeneity were deemed unnecessary.

In 3 ITSs, researchers reported on the effect of postpartum legislation on readmissions for treatment of jaundice^25,26,29  (Table 2). Meara et al²⁶  found a significant decrease in readmissions for jaundice after the minimum postnatal stay legislation, whereas Datar and Sood²⁵  found a nonsignificant reduction in the readmissions for jaundice in the second and third years after the legislation. Madden et al²⁷  found no difference in jaundice-related readmissions after introduction of the minimum stay law (Table 2).

In 2 ITSs, researchers reported on primary care use. Results from Madden et al²⁹  suggest that after adjustment for baseline trends, primary care use increased after implementation of the early discharge program, which slowly decreased by 1% per quarter (P < .01) after the minimum postnatal stay mandate (Table 2). Sievertsen and Wüst²⁸  found a significant increase in general practitioner (GP) contacts for infants who were discharged on the same day across all propensity groups (Table 2).

In 2 ITS studies, researchers reported on this outcome and found that attendances at emergency departments (EDs) decreased after introduction of the postnatal minimum stay mandate (Table 2).

No significant differences in the proportion of women breastfeeding at 48 hours postpartum were found in the meta-analysis of 3 trials that reported this outcome (RR: 1.05 [95% CI 0.99 to 1.11]; Supplemental Fig 5). No significant differences in the proportion of women breastfeeding between 1 and 2 months after birth were found in the 8 trials that reported this outcome^15–21,23  (pooled estimate RR: 1.01 [95% CI 0.94 to 1.09]) or 6 months after birth (RR: 1.18 [95% CI 0.98 to 1.43]; Supplemental Figs 6 and 7). The conclusion remained unchanged when Tan et al¹⁸  and Bayoumi et al¹⁵  were removed in subgroup analysis (RR: 1.05 [95% CI 0.98 to 1.12]).

In 2 ITSs, researchers reported on the proportion of infants breastfed before and after implementation of the law (Table 2). Madden et al²⁷  found no evidence of an effect on breastfeeding rates at 3 months. Sievertsen and Wüst²⁸  assigned a propensity score (based on whether the mother was married, unemployed, employed, and/or in education; higher education degree; and maternal age) and found that women in the lowest propensity score sample were less likely to breastfeed exclusively for at least 4 months if discharged on the day of birth (−0.311; P < .05), but the breastfeeding rates of women in the middle and highest propensity score groups were not affected (−0.213 [SE: 0.146] and −0.015 [SE: 0.244], respectively).

Researchers in only 1 trial assessed the proportion of women reporting infant feeding problems in the first 4 weeks after birth,¹⁶  showing significantly fewer (RR: 0.65 [95% CI 0.48 to 0.89]) in the first 4 weeks after birth in the standard LoS group.

The quality of individual studies was variable (Supplemental Figs 9 through 12). Risk of bias was assessed by using the EPOC criteria,⁷  which considers 4 domains: selection bias, performance bias, attrition bias, and reporting bias for both RCTs and ITSs (Supplemental Tables 5 and 6). The largest RCT with the greatest weight in the meta-analyses for the primary outcome was of reasonable quality, although attrition bias may have been an issue with >10% incomplete outcome data and differential noncompliance: 132 of 1890 participants refused to be discharged early in intervention group compared with 188 of 1896 who refused to be discharged at 72 hours.

Sensitivity analyses allowing for loss to follow-up in the RCTs was deemed inappropriate given the large proportion of participants lost to follow-up in many trials and considerable variation in how protocol violations were managed.

Overall, the quality of the 5 ITSs was good (Supplemental Figs 11 and 12). In all studies, the researchers reported outcomes that were described in the Methods section and had an intervention that did not affect data collection. In all ITS analyses, the slope of the intervention was prespecified, and authors acknowledged the potential effect other factors that may have coincided with the passage of the law (including changes to service mix, breastfeeding rates, physician awareness, and flu outbreaks).^24–27,29  Sievertsen and Wüst²⁸  also compared the trends in readmissions of primiparous women and women who had a cesarean delivery (who were not eligible for same-day discharge) to determine if additional policies (such as new medical routines at birth) had an effect on the outcomes.

This systematic review is, to our knowledge, the first to include evidence from both RCTs and ITSs with a predefined description of early discharge (<48 hours after vaginal birth and <96 hours after cesarean delivery) to assess the effect of a policy of early postnatal discharge and minimum LoS on health related outcomes. The pooled results of the 7 trials on infant readmission to the hospital indicated that more infants who were discharged early were readmitted compared with infants who had a >48 hours stay in hospital. The meta-analysis of ITSs, providing the next best available evidence on the subject, revealed that the US minimum stay law was an effective policy change, increasing postnatal LoS in hospital and providing evidence of a long-term reduction in infant readmission rates within 28 days of birth after this policy change. Because of lack of primary data, it was not possible to examine the effect of LoS on infant readmissions within 7 days.

In this review, we included 2 additional RCTs not included in the existing Cochrane review last updated in 2008.³  The authors of that review used evidence provided by study designs appropriate for policy intervention, both RCTs and ITSs, including those evaluating the impact of US federal and state legislation, and, therefore, provides a better understanding of the effect of postnatal LoS in both an experimental trial and naturalistic setting. By using EPOC criteria for study selection, a wider range of evidence was able to be included without compromising the quality of the findings, taking advantage of the evidence provided by good quality, well designed ITSs. This, in contrast to the RCTs, clearly demonstrated that interventions to institute a policy of early discharge actually resulted in increased early discharge of women and infants, allowing assessment of outcomes and, therefore, enhancing our knowledge of infant health outcomes in relation to early postnatal discharge policy in a “real life” setting.

This is the first study to conduct an ITS meta-analysis on this topic and provide an insight into the effect of federal and state law across several different state populations in the United States. Through our inclusion of these studies, we have also provided an understanding of the health-related outcomes for all infants, regardless of medical status or gestation at birth. A sensitive and broad search ensured that relevant evidence with any study design was included. Our review has also clearly defined early postnatal discharge, allowing more meaningful comparison across trials.

The limitations of this review also are reflected in poor trial quality and poor reporting. Despite the status of RCTs as the gold standard design for intervention studies, in this area, they have already been described as problematic because they feature high rates of postrandomization exclusions, crossover, and withdrawal. Many of the RCTs were low quality, lacking intention to treat analysis with resultant systematic differences between participants in the intervention and control group. In many trials, researchers did not adhere to current trial reporting standards, and, therefore, the findings from the RCT data should be treated with caution. We did not have any individual patient data, which may have provided more insight into subgroup analysis. Trials took place in several different countries where postnatal provision in the community may have varied considerably. The RCT analysis was insufficiently powered to explore early discharge without the addition of additional post discharge support in the community. There was also considerable clinical heterogeneity among the RCTs with regard to mode of delivery, with the largest RCT trial (providing 88% of the weight in the meta-analysis) only including women who had given birth via cesarean delivery. In light of the problematic participant dropout and crossover in the RCTs, the evidence from the ITS studies is particularly useful.

Nevertheless, findings from our review are the best evidence to date and, given problems with undertaking trials and other high-quality studies in this area, may well remain so. Differences found in meta-analysis of trial data for the neonatal readmission outcome are from one large study which only included women who had delivered via cesarean delivery: the findings must be treated with caution. There were no differences in outcomes related to maternal readmission or maternal depression, which might reflect insufficient power to detect these differences given relatively low incidence sample attrition. Breastfeeding rates were not measured in several studies. It was not possible to adequately report the effects of early postnatal discharge on primary care use. Across the trials, it was difficult to ascertain the proportion of mothers and infants who accessed primary care services, outpatient services, and unintentional injury and emergency care, and there were inconsistent definitions of primary care use and wide-ranging measurement methods. The data from ITSs were inconsistent, with researchers reporting both an increase and decrease in use after the postnatal mandate.

The definition of early discharge for this review (<48 hours for vaginal delivery and <96 hours for cesarean delivery) does not reflect the average length of postnatal stay for many high- and middle-income countries.¹  There was no evidence to support early discharges (<24 hours) from the hospital, a common practice internationally.³⁰  Although there is an assumption of reduced cost from earlier discharge, this may be off-set by increased costs associated with readmission or greater alternative care usage, and the cost-effectiveness of early discharge is unknown. Further research is needed to examine whether there are particular subgroups of infants who are most at risk for readmission if they and their mothers are discharged early or subgroups for whom early discharge is safe. Research on the impact of the short postnatal stays experienced in some settings is also needed. Research designs could build on the American Academy of Pediatrics’ Policy Statement on Hospital Stay for Healthy Term Newborn Infants³¹  to evaluate standardizing the discharge process through the use of predischarge checklists.

The use of state and federal law appeared effective in increasing the postnatal LoS for women and infants and resulted in a long-term reduction in the neonatal admission rate to hospital. Following the lead of the American Academy of Pediatrics statement,³¹  postnatal LoS policies should incorporate the needs of mothers and infants and not be led health services capacity or third party payers, ensuring that standardized systems are in place to ensure that women and infants are discharged at a time more appropriate to their needs.

Taken together, evidence from this meta-analysis and review of RCTs and ITSs of legislation mandating policies of a minimum stay postnatally and of minimum stay discharge policies has revealed that shorter postnatal stay in hospital (<48 hours after vaginal birth and <96 hours after cesarean delivery) is associated with increased infant readmissions to the hospital within 28 days of birth.

CBA

controlled before-after study

CI

confidence interval

EPOC

Effective Practice and Organisation of Care

GP

general practitioner

ITS

interrupted time series

LoS

length of stay

RCT

randomized controlled trial

RR

risk ratio