Objectives. To examine the proportion, geographic variation, and predictors of infant hospital readmission within 6 weeks of the postbirth discharge.
Methods. A cross-sectional, population-based study was conducted of all infants who were born from 1997 through 2001, linkable to the birth mother, and discharged alive from the hospital (N = 68 681) using hospital discharge files in the Canadian province of Manitoba. The following predictors of readmission were examined using logistic regression: preterm, low birth weight, neighborhood income, geographic location (the North, Rural South, and Urban areas of Winnipeg and Brandon), breastfeeding status, length of stay, maternal age, and type of delivery. Using 9 non-Winnipeg regions and 12 Winnipeg subregions, ecologic correlations (1-tailed Spearman) between newborn hospital readmission rates and the following were examined: 1) a region's overall health status, measured by the premature mortality rate (PMR), or death before aged 75 years and 2) a region's socioeconomic risk, using the Socio-Economic Factor Index (SEFI).
Results. The proportion of infants who were readmitted to the hospital at least once within 6 weeks of postbirth hospital discharge was 3.95%, with respiratory illness the leading cause (22.3% of readmissions). Risk of readmission was higher for infants who were born preterm (adjusted odds ratio [AOR]: 1.80; 95% confidence interval [CI]: 1.55–2.10), who were of the 3 lowest income quintiles (lowest: AOR: 2.02; 95% CI: 1.77–2.32; low: AOR: 1.48; 95% CI: 1.29–1.71; middle: AOR: 1.26; 95% CI: 1.08–1.47), who resided in the North (AOR: 1.85; 95% CI: 1.66–2.07) or Rural South (AOR: 1.25; 95% CI: 1.14–1.36), who were not breastfed (AOR: 1.32; 95% CI: 1.20–1.44), whose mother's age was 17 or younger (AOR: 1.30; 95% CI: 1.10–1.55), whose mother was 18 to 19 years of age (AOR: 1.25; 95% CI: 1.09–144), or who were born by cesarean section (AOR: 1.30; 95% CI: 1.19–1.43). Regional readmission rates were correlated with PMR (9 non-Winnipeg regions: r = 0.77 for PMR and r = 0.68 for SEFI; 12 Winnipeg Community Areas: r = 0.49 for PMR and r = 0.73 for SEFI).
Conclusions. Income and geography are strongly associated with newborn hospital readmission. Modifiable risk factors include increasing breastfeeding rates, decreasing cesarean section rates, and decreasing adolescent pregnancy rates (or increasing adolescent parental support), but these need additional study to establish causation.
Neonatal rehospitalization is an important issue, with significant morbidity and cost burdens.1,2 Previous reports in the Canadian province of Manitoba found the rate of neonatal rehospitalization to be 48 per 1000 live births during 1994 to 1998.3 A Wisconsin study reported 34 readmissions per 1000 term newborns within the first month of life.4 However, these rates cannot easily be translated into a percentage of liveborn infants who are readmitted, because they may include several readmissions of the same child. Danielson et al5 studied a cohort of healthy, term, vaginally delivered newborns in California and found that 3.02% of these infants were hospitalized at least once within the first month of life.
Much of the newborn readmission literature investigates associations with early discharge, with some showing increased risk with short length of stay1,5–12 and others no effect.13,14 Maternal age <20 years has also been associated with rehospitalization of normal birth weight infants,15 as well as being a first-time mother.12 Geographic variation in newborn readmission and possible relationships with regional health and socioeconomic status receive less attention, although low income has been identified as a risk factor.5,10 Manitoba, with a population of >1 million and a system of universal health insurance that theoretically removes financial barriers associated with hospitalization, provides a unique context in which to study such relationships.
The objectives of this population-based study were 1) to determine the proportion and risk factors of infants who are readmitted to the hospital within 6 weeks of postbirth discharge and 2) to examine ecologic associations of readmission rates with regional health status and socioeconomic status indicators.
This is a cross-sectional study of infants who were younger than 1 year; residing in Manitoba; and born between January 1, 1997, and December 31, 2001. Only infants who were discharged alive from the birth hospital stay were included in the analysis (N = 68 681). The data were derived from the deidentified hospital discharge files and vital statistics of the Population Health Research Data Repository stored at the Manitoba Centre for Health Policy. This contains hospital utilization information for all provincial residents, including dates of admission and separation, International Classification of Diseases, Ninth Revision, Clinical Modification diagnosis and procedure codes, birth date, and postal code. Of 70 308 birth records for 1997 through 2001, 70 072 (99.7%) infants and birth mothers were matched successfully. Removing duplicate records, 70 013 maternal/newborn matches remained; stillborn births and infants who did not leave the hospital alive were excluded, to yield 68 681 birth records for data analysis. Data management, computer programming, and statistical analyses used SAS.
The percentage of infants who were rehospitalized was analyzed by 11 geographic areas of Manitoba called regional health authorities (RHAs) and by 12 subregions of Winnipeg, the provincial capital, called Winnipeg community areas (WCAs). The RHAs in Fig 1 were also grouped into the following aggregate areas: the North (3 northern RHAs; n = 7436 infants); the Rural South (all southern RHAs excluding Winnipeg and Brandon; n = 22 231); and the 2 main Urban centers, Winnipeg (n = 36 349) and Brandon (n = 2665). Each area rate was compared with the overall provincial rate using Hotellings T2 method developed by Carriere and Roos16 for small area variation. To adjust for multiple comparison testing, we used 99% confidence intervals.
Our definition of hospital readmission does not separate the infants by their age at time of readmission, because infants were followed for 6 weeks postdischarge no matter what their age at the time of hospital discharge, truncated to exclude any infant who was kept in the hospital for 1 year or more. This definition differs from the “neonatal readmission” definition traditionally used by other studies, which confines the analysis to newborns who are readmitted within their first 28 days of life. “Neonatal readmission” is limited by excluding infants who stayed in hospital longer than 28 days. As well, there are inconsistent probabilities of hospital readmission for infants dependent on their length of stay. For example, an infant who is discharged on day 5 has a window of time of only 23 days in which to be readmitted as a neonate, compared with the 26-day window of an infant who is discharged on day 2.
Moreover, the traditional neonatal readmission rate captures the number of hospitalizations in the neonatal period per 1000 liveborn infants. One infant therefore could contribute >1 hospitalization within this period. In contrast, our definition of hospital readmission measures the proportion of liveborn infants who were discharged alive with 1 or more hospital stays taking place >1 day after discharge from the birth hospital stay. Direct transfers from the birth hospital to another hospital were not considered readmissions.
Cause of hospital readmission was based on the cause of first readmission. Predictors of at least 1 hospital readmission were determined through logistic regression modeling, using the explanatory variables of gestational age (preterm or not), region of residence (the 2 urban areas of Winnipeg and Brandon, North, and Rural South), type of feeding at hospital discharge (breastfed or not), income quintile grouping, type of delivery (vaginal or cesarean section), birth weight (low, high, or other), maternal age (17 or younger, 18–19, and 20 or older), and maternal length of stay (<2 days or not).
Preterm was defined as gestational age <37 weeks; low birth was defined as weight <2500 g. Breastfeeding initiation rates were approximated using the hospital discharge abstracts, where feeding type of newborns is recorded on discharge as “breast,” “both breast and artificial,” or “artificial.” Grouping together “breast” and “both breast and artificial” to produce “any breastfeeding” is the most valid, because it has been noted that different institutions use different amounts of supplement as the cutoff for coding “both” versus “only breastfed.” In the hospital discharge abstracts, 1.95% of charts were missing feeding data provincially. The discharge abstracts have demonstrated a high degree of validity (99%) when compared with maternal report on type of infant feeding initiated in the hospital.17
Residents of Manitoba were placed into urban (Winnipeg and Brandon) or rural (all other RHAs) income quintiles on the basis of their postal codes. Postal codes were sorted separately by average household income value at the Enumeration Area level, based on 1996 census data. Approximately 20% of the population in both rural and urban Manitoba was allocated to each quintile. Individual-level income is not available in the Population Health Research Data Repository. However, previous studies have used this ecologic measure of income,18,19 and there is a moderate correlation (0.41–0.45) between enumeration area quintile and individual quintile assignment. Using this ecologic measure rather than an individual income also points to the fact that the individual may be just as affected by neighborhood as by individual income.20
For the 11 non-Winnipeg RHAs and the 12 WCAs, regional rates of hospital readmission were correlated with the regional premature mortality rate (PMR) and a composite measure called the Socio-Economic Factor Index (SEFI), using nonparametric 1-tailed Spearman rank correlation test, which correlates rank order rather than actual data. This protects against effects of outliers and nonnormal data distributions. Because of small regional population (n = 1032) for the Churchill RHA, with resulting highly variable rates, this region was excluded in correlations and scatter plots. A weighted regression that included Churchill yielded similar correlation results, but normality assumptions were questionable. PMR is an age- and gender-adjusted rate of death before the age of 75 years, standardized to the age and gender structure of the Manitoba population in 1996, and is often considered a robust regional overall health status measure.21 PMRs for the geographic divisions of the province were based on 5-year averages to ensure stability of the measure. SEFI is based on an index derived from municipal-level census measures for labor force participation of women, age-dependency ratio, percentage of single-parent households, percentage of female single-parent households, and 2 aggregated factors representing unemployment and education.21 The lower the SEFI, the more favorable the socioeconomic conditions.
Of the 68 681 newborns from 1997 through 2001, 2716 were readmitted to the hospital within 6 weeks postbirth discharge. Figure 2 illustrates that 3.95% ± 0.15 of all Manitoba infants were readmitted at least once, with the highest proportion in the North (6.59% ± 0.56), compared with Rural South (4.07% ± 0.26), Winnipeg (3.35% ± 0.18), and Brandon (3.94% ± 0.74). The timing of hospital readmissions did not seem to be area specific, with readmissions occurring primarily within the first 2 weeks after discharge. The leading provincial cause of hospital readmission was respiratory illness (22.3%), with the highest proportion in the North (Table 1).
A model for risk of hospital readmissions within 6 weeks postbirth discharge was significant (likelihood ratio χ2 = 566, df = 17, P < .0001), with the following significant variables: preterm, not being breastfed, born by cesarean section, maternal age <20 years, residing in a low- to middle-income area, and residing in the North or Rural South areas (see Table 2 for odds ratios [ORs]). Maternal length of stay was not significant. Low birth weight was significant in the univariate model but was not significant when adjustment was made for such overlapping risk factors as preterm birth.
In the ecologic analyses, hospital readmissions were significantly associated with regional PMR and SEFI values. For the 9 non-Winnipeg RHAs, with Churchill excluded because of small sample size, the Spearman correlation coefficients (1-tailed) were r = 0.77 (P < .008) for PMR and r = 0.68 (P < .02) for SEFI. For the 12 WCAs, the correlation coefficients (1-tailed) were r = 0.49 (P < .05) for PMR and r = 0.73 (P < .004) for SEFI (Figs 3 and 4).
Hospital readmission of newborns within 6 weeks of postbirth discharge is a problem for the infant, the family, and the health care system and must be examined from a perspective of preventing illness. At the individual level, preterm birth, not being breastfed at hospital discharge, cesarean section birth, having an adolescent mother, being a resident in census areas in the 3 lowest income quintiles, and being a resident outside the urban centers of Winnipeg and Brandon all were associated with elevated risk of hospital readmission within 6 weeks postbirth discharge, even after adjusting for the co-presence of other risk factors.
Table 3 includes data for the aggregated areas of Winnipeg, Brandon, and Rural South and North areas, derived from 2 Manitoba Centre for Health Policy reports.22,23 In this study, the highest hospital readmission rate occurred in the North. Although preterm and low birth weight rates are similar to the provincial average, teen pregnancy rates in the North are double the provincial rate (121.4, 95% confidence interval [CI]: 114.8–128.4; vs 61.4; 95% CI: 60.0–62.8 births per 1000). Breastfeeding rates in the North are also lower, at 64.9% (95% CI: 63.5–66.4%) of newborns breastfed on hospital discharge in the North compared with 79.7% (95% CI: 79.3–80.1%) provincially. However, the geographic area of the North, even after controlling for maternal age, breastfeeding, preterm, low birth weight, and income, shows an independent effect in the logistic regression analysis (Table 2). This particular area of Manitoba is at greatest socioeconomic risk and has the poorest overall regional health status (ie, the highest premature mortality rate) in the province. The evidence of the North's geographically independent effect may possibly be found in both individual and area effects. For example, the elevated regional smoking rates could very well contribute to significantly poorer health status and respiratory infection of the newborn. Other contributors may be more subtle, such as maternal stress and poorer living conditions of the area. Moreover, in the remote areas of the North, a health care provider may well err on the side of caution when deciding on readmitting an ill infant, because of the vast distance (often by airplane) to acute-care facilities.
What are the modifiable risk factors in the rehospitalization of infants? The effect of being an adolescent mother possibly could be modified through supportive programming. One randomized trial of a public health nurse pre- and postnatal program for adolescent parents found significantly fewer total days of infant hospitalization and rehospitalization during the first 6 weeks of life compared with the control group of regular public health nurse contact.24 During frequent postnatal visits (17 visits from third trimester through the first year after birth), mothers were given information on health, family planning, maternal role (including infant care), life skills, and social support.
Breastfeeding was also found to be protective against readmission, most likely associated with the major cause of readmission being respiratory illness. Breastfeeding has been shown to afford protection in a dose-response relationship against such infant illnesses as gastrointestinal and respiratory infections,25–29 even in smoking households.30,31
Given the major morbidity and cost burdens associated with breastfeeding-preventable neonatal illnesses, provincial health care providers and decision makers need to support regional programs to promote breastfeeding, especially to high-risk groups. Breastfeeding initiation rates vary widely by geography, socioeconomic status, and maternal age; women in regions that have the least healthy population and the greatest socioeconomic risk, adolescent mothers, and women in the lowest income groups are the least likely to initiate breastfeeding both in urban and in rural areas.32 Individual breastfeeding support in the form of peer support programs has been shown to be effective in increasing breastfeeding initiation and duration and decreasing morbidity in both developing and developed nation contexts.25,33–36
Short stays of <2 days were not associated with increased risk of rehospitalization (OR: 1.05; 95% CI: 0.93–1.18; not significant). This finding contradicts Liu et al,9 where length of stay <30 hours postbirth was associated with greater risk of readmission within 7, 14, and 28 days of life, controlling for parity and maternal age (OR: 1.12–1.28). However, the high adjusted OR associated with preterm birth (OR: 1.80; 95% CI: 1.55–2.10) seems to support the assertion of Marbella et al4 that efforts to reduce preterm births may have a great impact on neonatal health and health care costs. That being said, there is little variation in preterm birth rate by region (Table 3). According to the Manitoba Perinatal Health Surveillance Report, the provincial preterm birth rate in the 1997/1998 fiscal year was 7.6% but 9.5% among women who smoked.3 This indicates that a strategy of smoking cessation may possibly reduce preterm rates but also reduce other risk factors for respiratory infection of the newborn beyond simply birth weight.
The cross-sectional nature of the analyses, as well as the ecologic correlations, must be used with caution, with the limitation that risk factors for infant hospital readmission are associations but not necessarily causal. As well, this study is limited by the use of person-specific information available in the Manitoba hospital discharge abstracts (with the exception of a small-area average household income attributed to the individual), where there is no available individual information on lifestyle issues such as smoking, stress, household condition, or any number of variables that may have an impact on infant readmission. Such data are available only on national survey samples, not in a “linked” format to our existing universal database. Moreover, these national surveys do not include “on-reserve” First Nations communities, eliminating a high proportion of the Aboriginal people (28% of the population in the North and 48% in the RHA of Burntwood live on-reserve37).
Factors that affect the readmission rates of infants within 6 weeks postbirth discharge need additional study to determine whether systematic programs in the prenatal or postnatal period could reduce rates. Such targeted interventions as support for adolescent parents, breastfeeding support, smoking cessation, and reduction of cesarean section rates need population-based evaluation studies to determine their regional effectiveness. Moreover, it is important to note that even after adjusting for individual risk factors (eg, preterm birth, breastfeeding status), both the region of residence (OR: 1.85; 95% CI: 1.66–2.07) and neighborhood income (OR: 2.02; 95% CI: 1.77–2.32 for lowest quintile) have an impact on the risk of readmission and indeed are the greatest risk factors. Ecologic associations of readmission rates and overall regional health and socioeconomic status measures underscore the “big picture” risk factors. The challenge is to examine how approaches that are designed to reduce socioeconomic disparity, possibly through social policy decisions and economic support, ultimately affect infant hospital readmissions.
This work was supported as part of a project on child health in Manitoba, one of several projects undertaken each year by the Manitoba Centre for Health Policy under contract to Manitoba Health. The results and conclusions are those of the authors, and no official endorsement by Manitoba Health was intended or should be inferred. The full report, “Assessing the Health of Children in Manitoba: A Population-Based Study,” is available from the Manitoba Centre for Health Policy, University of Manitoba (available at www.umanitoba.ca/mchp/centres). The principal author (Dr Patricia J. Martens) holds a New Investigator Award by the Canadian Institutes of Health Research (CIHR), as well as a CIHR Community Alliances for Health Research Program grant.
We thank Sandra Peterson for programming assistance in the original report, as well as Dr Marni Brownell for helpful suggestions in the area of newborn readmissions and in adolescent parenting issues.
- ↵Manitoba Health. Public Health Branch Epidemiology Unit Perinatal Project Team. Manitoba Perinatal Health Surveillance Report 1989–1998. Winnipeg, Manitoba, Canada: Manitoba Health; 2000
- ↵Marbella AM, Chetty VK, Layde PM. Neonatal hospital lengths of stay, readmissions, and charges. Pediatrics.1998;101 :32– 36
- ↵Danielsen B, Castles AG, Damberg CL, et al. Newborn discharge timing and readmissions: California, 1992–1995. Pediatrics.2000;106 :31– 39
- Malkin JD, Broder MS, Keeler E. Do longer postpartum stays reduce newborn readmissions? Analysis using instrumental variables. Health Serv Res.2002;35 :1071– 1091
- ↵Hall RT, Simon S, Smith MT. Readmission of breastfed infants in the first 2 weeks of life. J Perinatol.2002;20 :432– 437
- ↵Escobar GJ, Joffe S, Gardner MN, et al. Rehospitalization in the first two weeks after discharge from the neonatal intensive care unit. Pediatrics.1999;104(1) . Available at: www.pediatrics.org/cgi/content/full/104/1/e2
- ↵Martens PJ. Breastfeeding Choice and Duration: A Prospective Study of Women and Infants in Four Southern Manitoba First Nations Communities. Master's Thesis. Winnipeg, Manitoba, Canada: University of Manitoba; 1994
- ↵Smith GD, Hart CL, Watt G, Hole D, Hawthorne V. Individual social class, area-based deprivation, cardiovascular disease risk factors, and mortality: the Renfrew and Paisley study. J Epidemiol Community Health.1998;52 :399– 405
- ↵Martens PJ, Fransoo R, The Need to Know Team, et al. The Manitoba RHA Indicators Atlas: Population-Based Comparison of Health and Health Care Use. Winnipeg, Manitoba, Canada: Manitoba Centre for Health Policy; 2003
- ↵Brownell M, Martens P, Kozyrskyj A, et al. Assessing the Health of Children in Manitoba: A Population-Based Study. Winnipeg, Manitoba, Canada: Manitoba Centre for Health Policy and Evaluation; 2001
- Oddy WH, Sly PD, de Klerk NH, et al. Breast feeding and respiratory morbidity in infancy: a birth cohort study. Arch Dis Child.2003;88 :224– 228
- Scariati PD, Grummer-Strawn LM, Fein SB. A longitudinal analysis of infant morbidity and the extent of breastfeeding in the United States. Pediatrics.1997;99(6) . Available at: www.pediatrics.org/cgi/content/full/99/6/e5
- ↵Wilson AC, Forsyth JS, Greene SA, et al. Relation of infant diet to childhood health: seven year follow up of cohort of children in Dundee infant feeding study. BMJ.1998;316 :21– 25
- ↵Nafstad P, Jaakkola JJK, Hagen JA, et al. Breastfeeding, maternal smoking and lower respiratory tract infections. Eur Respir J.1996;9 :2623– 2629
- ↵Martens PJ, Derksen S, Mayer T, Walld R. Being born in Manitoba: a look at perinatal health issues. Can J Public Health.2002;93 :S33– S38
- ↵Martens PJ. Increasing breastfeeding initiation and duration at a com-munity level: an evaluation of Sagkeeng First Nation's community health nurse and peer counselor programs. J Hum Lact.2002;18 :236– 246
- Dennis CL, Hodnett E, Gallop R, Chalmers B. The effect of peer support on breast-feeding duration among primiparous women: a randomized controlled trial. CMAJ.2002;166 :21– 28
- ↵Martens P, Bond R, Jebamani L, et al. The Health and Health Care Use of Registered First Nations People Living in Manitoba: A Population-Based Study. Winnipeg, Manitoba, Canada: Manitoba Centre for Health Policy; 2002
- Copyright © 2004 by the American Academy of Pediatrics