Rehospitalisation Risk Factors and Rates in Preterm Babies: A Systematic Review and Meta-Analysis

Objective: To systematically review evidence of risk factors and rates for rehospitalisations within one month of discharge for babies born at <37 weeks gestation. Design: Systematic and meta-analysis Study Selection: Inclusion criteria were studies published in English or French between 01 January 2000 to 31 March 2019, recruiting from the year 2000 onwards evaluating risk factors for rehospitalisation within one month of discharge in preterm babies. Two reviewers independently selected relevant studies, extracted study details, baseline characteristics and results of risk factor analyses. Results: Across 14 included studies, ve studied babies of <37 weeks gestation, seven studied late preterm babies (34-36 weeks gestation), and two studied very to moderate preterm babies (<34 weeks gestation). Important risk factors were low birth weight, respiratory morbidity, male sex and lower socioeconomic status in <37 week babies, and shorter length of stay among late preterm babies. Pooled rehospitalisation rates were 4.3% (95% CI 1.9-9.7) in <37 week babies and 6.6% (95% CI 3.2-13.4) in late preterm babies. There was high heterogeneity in risk factors included in analyses and studies often lacked clarity on variable measurement and confounder adjustment. Conclusion: We found evidence for clinical and socioeconomic risk factors, high heterogeneity and important limitations. Limitations included a lack of breadth in both the gestational age ranges and risk factors studied, as well as lack of clarity around variable measurement and confounder adjustment


Introduction
Babies born at < 37 completed weeks gestation are more vulnerable than full-term babies. Outcomes in preterm babies are closely linked to gestational age (GA) at delivery [1], and babies born before 32 weeks, and particularly those before 27 weeks gestation, experience the greatest risk of short and long term mortality and morbidity. [2][3][4][5][6] Advances in care have improved outcomes among preterm babies [7,8], although this increased survival has at times been accompanied by more being discharged with serious morbidities [8][9][10].
Rehospitalisations temporally close to discharge can provide insight into care and discharge quality. [11,12] Measures of rehospitalisation are used as indicators of quality and utilisation by health systems such as Medicaid in the United States [13,14] and the National Health Service in the United Kingdom [15]. Early rehospitalisations are burdensome and costly, particularly among vulnerable preterm populations. [16,17] There has been signi cant amounts of research into risk factors for early rehospitalisation in preterm babies, but to the best of our knowledge there are no systematic reviews of risk factors for early rehospitalisation. Improving our understanding of rehospitalisation risk factors could inform clinical, discharge and policy decisions. We sought to review the evidence pertaining to the risk factors for, and likelihood of, early rehospitalisation among preterm babies. Our primary objective was to conduct a systematic review of risk factors for early rehospitalisation in preterm babies, with a secondary objective of analysing rates of early rehospitalisation across the included studies.

Methods
A systematic review and meta-analysis of risk factors for rehospitalisation (within one month of discharge) in babies born at < 37 weeks gestation was performed and reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines. [18] (Online supplemental1) Two reviewers (RAR and ASM) independently conducted study selection, data extraction and quality assessment. Disagreements were resolved through further discussion or referral to a third author (BK). The review protocol was registered with the International Prospective Register of Systematic Reviews (PROSPERO); CRD42018086549.

Databases and search strategy
PubMed (including MEDLINE and life science journals), Web of Science and reference lists of included articles were searched for studies published between January 01, 2000 and March 31, 2019. Search terms for titles and abstracts included: ('preterm', 'premature', 'near term', 'neonate', 'gestation', 'birth weight') and ('readmission', 'rehospitalisation', 'readmitted') and ('risk factor', 'risk', 'determinant', 'cause', 'variables'). Medical Subject Headings (MeSH) and wildcard terms were included when available. The search strategy and term were developed in consultation with a systematic review expert. Full details are provided in Online supplemental 2.

Inclusion and exclusion criteria
We included observational and experimental studies published in English or French, as well as abstracts. Study populations were either exclusively preterm (babies born at < 37 weeks GA) or -in studies including both full-term and preterm babies in their samples -included an analysis of a sample subset, containing preterm babies only. Included studies investigated risk factors for rehospitalisations occurring within a maximum of one month of discharge. We excluded studies recruiting patients before the year 2000, those with only full-terms (≥ 37 weeks GA) or where the preterm population could not be distinguished, and those investigating interventions speci c to their local environment.

Data extraction
Study characteristics (concerning design, sampling and analyses) and descriptive statistics were extracted from eligible articles. All data relating to rehospitalisation risk factor analyses were extracted. Rates of early rehospitalisation were also extracted from all included studies. In cases where rate data were not provided in articles, data were requested directly from authors.

Quality assessment
Quality of analyses relating to the identi cation of risk factors for rehospitalisation were independently evaluated by both reviewers using the relevant tools from the Critical Appraisal Skills Programme (CASP). [19] Analysis Due to heterogeneity an overall meta-analysis of risk factors was not possible and a narrative synthesis was performed instead. The synthesis was grouped according to the GA categories studied. Studies were grouped into logical categories whilst also maximising the number of studies in each. Groupings were de ned in this study as: (1) all preterm babies (studies of < 37 weeks GA); (2) late preterm babies (studies of 34-36 weeks GA); (3) very to moderate preterm babies (studies of < 34 weeks GA or very low birth weight (VLBW) babies).
Two meta-analyses of rehospitalisation rates were performed: the rst related to studies of all preterm babies (< 37 weeks GA) and the second focussed on studies of babies born at 34-36 weeks gestation; insu cient data were available for other GA groupings. Studies investigating rehospitalisation within follow-up periods signi cantly shorter than one month (15 or 7 days for example) were insu cient in number to be included in a separate meta-analysis. Study rates were combined using a random effects model and presented in forest plots. The pooled proportion was estimated using logit transformation with the inverse variance method. Heterogeneity between studies was assessed using the I 2 test statistic. [20] All analyses used R version 3.4.2 [21] and the 'meta' package [22].

Description of included studies
The initial search returned 1,580 records, 169 underwent full text review, with 156 excluded. Thirteen records met inclusion criteria and the manual search of reference lists for included articles identi ed one additional study [14]. Full details can be found in Fig. 1. Extracted study details are presented in Table 1 and population characteristics in Online supplemental 3. Details of quality assessment are shown in Table 2 and Online supplemental 4. Of the four cohort studies, Jensen et al 23  In studies that used health system or insurance databases 23,25,30 details of how data were extracted, coded and extent of validation were not provided. Two studies did not clearly de ne how GA was established or measured 26,30 In addition, Jensen et al 23 acknowledged that testing for their key risk factor of car seat tolerance screening may vary widely between centres and they assumed that initial screenings were all performed in an infant car seat. Further lack of clarity over the identi cation, selection and, inclusion of confounders was present in Jensen et al 23

Treatment & care
All four studies investigating length of stay found that shorter rather than longer length of stay increased RH risk. 14,24,29,34 Morbidities Oltman et al 27 found preterm babies with high levels of certain metabolites C16:1, C14:1 and C3, and high tyrosine to ornithine ratios had an increased risk of RH.

Maternal
No studies investigating this population of preterm babies included maternal variables in their analyses.

Sociodemographic
No studies found an effect from demographic and environmental variables.

Discussion
In this systematic review of the literature on risk factors for, and likelihood of, early rehospitalisation in preterm babies we found 14, albeit heterogeneous articles. These articles were heterogeneous in terms of populations, gestational ages and risk factors in analyses. Several large studies found that sub-optimal birth weight, respiratory morbidities, being male and lower socioeconomic status were risk factors for rehospitalisation among babies of < 37 week gestation. There was also consistent evidence that shorter length of stay increased the risk of rehospitalisation in late preterm babies, though this evidence was of limited quality due to inadequate control for confounders and poor reporting of missingness and loss to follow-up. Evidence pertaining to babies born at < 34 weeks gestation was less abundant, comprising predominantly small samples with limited detail concerning confounder adjustment and missing data. Meta-analysis of rehospitalisation rates is di cult to interpret given marked heterogeneity between studies.

Evidence for risk factors of early rehospitalisation
In preterm babies as a whole, lower birth weight was consistently a risk factor for rehospitalisation [25,26,30]. This nding was most likely in part due to the fact that birth weight is correlated with gestational age, which is a key determinant of physiological immaturity and outcome in newborns. Conversely, Schell et al [26] found that being SGA was associated with a lower risk of rehospitalisation, perhaps indicating that babies bene t from the more intensive management potentially afforded to SGA babies. Evidence of a role for respiratory morbidity (apnea [25] and lung disease [30]) in rehospitalisation was shown in two large studies, was unsurprising given the strong association between lower gestational age and respiratory system immaturity. [4,36] Two studies [25,26] found Medicaid recipients had an increased rehospitalisation risk and Tseng et al [30] found lower insurance premium payments increased risk. These ndings regarding contextual factors related to socioeconomic status are supported by the literature [37][38][39] and present challenges for clinicians, as such factors may not b e possible to in uence. Being male was a risk factor for rehospitalisation among preterm babies (< 37 weeks gestation) in a US [25] and Taiwanese [30] study, re ecting the well established male disadvantage [40,41], likely accounted for by physiological, hormonal and developmental factors. [41] Among late preterm babies, multiple studies [14,24,29,34] found shorter length of stay to be a risk factor. However, the failure of three of the studies [14,24,29] to adjust for environmental factors such as socioeconomic status or maternal variables hinders interpretation of these results.

Rates of early rehospitalisation
The pooled rate of rehospitalisation within 31-days was 4.3% in < 37 week babies and 6.6% for late preterms. Differences in populations, methodology and clinical practice may account for some of the variations across studies. The highest rates came from studies from Taiwan [30] and India [33] respectively, whilst the lowest rates were seen in US studies [14,23]. This disparity could in part be explained by differences in discharge criteria, preparation of families, and follow-up between regions. The lower rate seen in Young et al [14] versus Harron et al [29] is somewhat surprising as Harron et al [29] considered only unplanned rehospitalisations while Young et al [14] considered all rehospitalisations, including planned admissions for surgery or investigations. Despite this, the discrepancy might be explained by more conservative discharge practices in the United States. Establishing whether differences in rate were due to divergence in sample risk pro les was not possible as the reporting of relevant baseline characteristics was not comprehensive in Harron et al [29], Mallick et al [33] nor Young et al [14]. Moreover, we also acknowledge that the use of interview to ascertain rehospitalisation status by several studies [26,28,[31][32][33] could have introduced recall bias, leading studies such as Mallick et al [33] to underestimate rehospitalisation rates.

Strengths and limitations of the eligible studies
Almost all studies recruited over a period of at least one year, thus limiting the in uence of seasonality on outcome measures. Multiple studies used large samples, recruited from regional or national databases -thus improving representativeness -and used statistical methods that facilitated adjustment and produced useful measures to quantify effect sizes.
There were many limitations across the studies. Recruitment from single centres and corresponding small sample sizes was relatively common, that could have resulted in limited power to detect associations with potential risk factors [20] and a higher likelihood that samples were not representative of the underlying populations. Many studies used electronic health data, insurance databases or medical records. These are vulnerable to biases related to missing data and systematic measurement error. As such, it is important that such administrative data are validated [42], yet only Harron et al [29] used data that was clearly routinely validated [43]. Other studies used interviews to establish the outcome and these are vulnerable to recall and social-desirability bias, as well as misunderstanding of questions on behalf of the interviewee.
Insu cient adjustment for confounders or effect modi ers affected over half [14, 23, 24, 27-29, 32, 33] of the fourteen studies and gestational age, birth weight and socioeconomic status were too infrequently adjusted for in particular. The processes used to select confounders was rarely described in the studies. This made comparison across studies challenging and reduced the likelihood that measures were quantifying the true or even the same effect. Under half of the studies reported the proportion of loss to follow-up [25,28,[30][31][32][33], and just one [32] compared eligible babies to those lost to follow-up. Fewer than half of studies reported details of missing data [24-26, 29, 30] or conducted relevant analyses [26,29].
Consequently, it was di cult to establish the potential role of attrition bias.

Strengths and limitations of this review
This systematic review provides novel results, synthesising the scienti c literature as it relates to risk factors for, and the likelihood of, early rehospitalisation in samples of preterm babies. We followed PRISMA guidelines [18], and the development of the search strategy was made in consultation with an expert in systematic reviews. We consulted multiple large databases, included abstracts, and conducted additional reference searches. We believe our exclusion of studies of speci c local interventions and pre-year 2000 populations ensured the broad relevance and generalisability of this review to modern standards of neonatal care, across contexts. Our use of the CASP tools is an added strength given its comprehensive scope and lack of reliance upon potentially arbitrary and reductive aggregate article scores. [20] We acknowledge that we may have missed some studies given our inclusion of only English and French articles. In addition, heterogeneity in GA categories complicated synthesis of study ndings and the size of the evidence-base data related to each GA category. Categorisation of GA in the literature and this review's synthesis could be deemed arbitrary as the effects of prematurity are likely present on a continuum. Study heterogeneity made pooling of risk factor effects impractical and meant we could not effectively assess publication bias. We also assumed that studies using follow-up periods of 31 days or less were measuring the same outcome. Though most studies used a follow-up of approximately one month, we acknowledge that babies rehospitalised within one week may have different risk factor pro les, complicating interpretation of this review.

Conclusions
Among babies of < 37 weeks gestation, sub-optimal birth weight respiratory morbidities, being male and lower socioeconomic status were associated with early rehospitalisation. Shorter length of stay was a consistent risk factor for early rehospitalisation in studies of late preterm babies. This review also highlights a lack of high quality research and substantial heterogeneity among studies addressing late (34-36 weeks GA) and very to moderate preterm (< 34 weeks GA) babies in particular. However, despite limitations in the literature, there is little to suggest that clinicians are not broadly managing the discharge and follow-up of preterm babies effectively. Future research should consider wider ranges of gestational age (particularly < 34 weeks gestation) and risk factors, while adjusting more comprehensively for confounding. Strengthening the evidence-base on early rehospitalisation could improve clinical decision making, thus reducing healthcare burdens and improving outcomes for preterm babies. Not applicable. Systematic review and meta-analysis study design.

Consent for publication
Not applicable.

Availability of data and materials
All the data collected in this study are are provided in either the main article or supplementary materials.

Competing interests
The authors declare that they have no competing interests. Author's contributions Robert A. Reed and Dr Andrei S. Morgan conceptualized and designed the study, collected data, carried out the initial analyses and drafted, reviewed and revised the manuscript. Dr Babak Khoshnood, Dr Pierre-Yves Ancel and Dr Jennifer Zeitlin conceptualized and designed the study and drafted, reviewed and revised the manuscript. Dr Agnès Dechartres designed the study and reviewed and revised the manuscript. All authors approved the nal manuscript as submitted and agree to be accountable for all aspects of the work.