Effects of brief alcohol interventions in Swedish maternity care

A large body of research documents the importance of early life conditions for the health and human capital formation of children. The detrimental effects of alcohol exposure in utero are well documented, and therefore identifying effective methods for preventing harmful maternal alcohol consumption is of great importance. We exploit the stepwise introduction of alcohol screening and brief interventions at Swedish antenatal clinics, to evaluate the causal effect of enhanced alcohol prevention on infant health using a difference-in-differences strategy. We find that the program improves infant health measured by prescription of pharmaceutical drugs and hospitalizations during the child’s first year of life. The results suggest that effects are likely driven by changes in maternal behavior after the first trimester and seem to extend beyond the birth of the child. Keywords: Alcohol prevention; Brief intervention; AUDIT; Antenatal care; Child health JEL-codes: I12; I18 a We are grateful for comments and suggestions from Caroline Hall, Kristiina Huttunen, Erica Lindahl, Kerstin Petersson, Stephanie von Hinke Kessler Scholder and participants at the Family and Education Workshop 2016, Puerto Rico, Workshop Health and the Labour Market 2016 at Aarhus university, Nordic Health Economic Study Group Meeting in Uppsala 2015, 10th Nordic Summer Institute in Labor Economics in Uppsala, SOLE meeting 2016 in Seattle and seminars at Stockholm School of Economics, Linnaeus University and Health Economic Forum at Uppsala University. b Institute for Evaluation of Labour Market and Education Policy, [email protected] c Department of Economics, Uppsala University, [email protected] d Institute for Evaluation of Labour Market and Education Policy and UCLS, Uppsala University, [email protected] e Department of Economics, UCLS and UCFS Uppsala University and Linnaeus University, , [email protected] Work in progress -­ do not quote 2 IFAU Sober mom, healthy baby? Table of contents 1 Introduction ......................................................................................................... 3 2 Prenatal health and alcohol exposure ................................................................. 7 3 Antenatal Care, Screening and Brief Interventions ............................................ 9 4 Empirical strategy ............................................................................................. 12 4.1 Expected effects of the program ....................................................................... 15 5 Data ................................................................................................................... 17 5.1 Study population and screening ........................................................................ 17 5.2 Child health outcomes ...................................................................................... 19 5.3 Descriptive statistics ......................................................................................... 20 5.4 AUDIT scores, maternal characteristics, behaviors and child outcomes ......... 21 6 Results ............................................................................................................... 22 6.1 The effect of the program on child health ........................................................ 22 6.2 Which health conditions are affected? .............................................................. 24 6.3 Heterogeous effects .......................................................................................... 25 6.4 Sex differences .................................................................................................. 27 6.5 Socio-economic outcomes of parents ............................................................... 28 6.6 Robustness of results ........................................................................................ 29 7 Effects of the program on pregnant women’s behavior using survey data ....... 33 8 Conclusion ........................................................................................................ 36 References ....................................................................................................................... 38 Appendix A ..................................................................................................................... 46 Appendix B. AUDIT-questionnaire ................................................................................ 51 Work in progress -­ do not quote IFAU Sober mom, healthy baby? 3 1 Introduction Public interventions and recommendations concerning expecting women’s alcohol consumption have long been part of national strategies to promote maternal and child health. This has been motivated by the insight that the fetus is not protected from harm in utero and by evidence of negative effects of alcohol exposure (McBride, 1961; Von Lenz and Knapp, 1962; Jones, Smith, Ulleland and Streissguth, 1973; Barker 1990). Ambiguous findings regarding the effects of moderate alcohol consumption during pregnancy have however lead to a questioning of strict recommendations to completely abstain from alcohol (see for example Oster; 2013), and pregnant women do not always follow the recommendations. In spite of strict recommendations in Sweden, Göransson et al (2003) find that about 30 percent of pregnant women reported using alcohol regularly, in an anonymous survey. Barry et al (2009) report much lower figures for the US: 10-12 percent of pregnant women report drinking at all. Yet, this is of concern in view of a growing recent literature in economics showing that alcohol exposure in utero has causal adverse effects on health and human capital (see e.g. Wüst, 2010; Zhang, 2010; von Hinke et al., 2014; Nilsson, 2015); in particular since Wüst and von Hinke are able to demonstrate that the ambiguous impact on child health of maternal wine or moderate alcohol consumption disappear when selection effects are accounted for. In a report of the US National Task Force on Fetal Alcohol Syndrome and Fetal Alcohol Effect it is concluded that research on the effectiveness of universal prevention interventions to reduce alcohol related pregnancies or fetal alcohol spectrum disorders is insufficient, though Screening and Brief interventions are mentioned as promising strategies (Barry et al, 2009). Hence, it is of great importance to identify effective methods for preventing harmful fetal alcohol exposure, and more generally to find interventions that improve child health. It is also important to understand how enhanced 6 Prenatal exposure to alcohol is identified as an important preventable cause of mental retardation with large medical and social costs (Abel and Sokol, 1987; West and Blake, 2005). The insight that the fetus is not protected from harm in utero has gained recognition since the 1960s. The documentation of the severe side effects of Thalidomide in the 1960’s (McBride, 1961; Von Lenz and Knapp, 1962) and of adverse effects of alcoholism in the early 1970’s (Jones, Smith, Ulleland and Streissguth, 1973) was important for establishing the vulnerability of the fetus. These and other findings lead Barker (1990) to formulate the Fetal origins hypothesis, which is discussed at length in Almond and Currie (2011). There is now a large empirical literature documenting effects on health and human capital of fetal exposure to toxic substances (Chay and Greenstone 2003; Almond, Edlund and Palme, 2009; Currie, Niedell and Schmeider, 2009; Currie, Greenstone and Moretti, 2011; Currie and Walker 2011; Black et al. 2013), maternal health shocks (Almond 2006), malnutrition (Lindeboom, Portrait and van der Berg 2010; Almond and Mazumder, 2011; Doblehammer, van der Berg and Lumey 2011;), maternal stress (Currie and Rossin-Slater, 2013; Lindo, 2011), economic conditions (van der Berg, Lindeboom and Portrait, 2006; van der Berg, Doblhammer and Christensen 2011), and alcohol (Wüst, 2010; Zhang, 2010; von Hinke et al., 2014; Nilsson, 2015). Work in progress -­ do not quote 4 IFAU Sober mom, healthy baby? preventive interventions against health hazards in utero affect health and early development of children. The contribution of this paper is to do just that. We exploit regional time variation 2004-2009 in the introduction of the Swedish Risk Drinking project in antenatal care. This is a screening and brief intervention (BI) program for alcohol in Swedish antenatal clinics from 2004 to analyze the effects of enhanced alcohol prevention on child health and maternal behavior during the first years of life. The program consists of three parts: (i) screening for risky alcohol consumption in gestation week 8-12 using the Alcohol Use Disorder Identification Test (AUDIT) instrument as a pedagogic tool to screen and inform about risks; (ii) using Motivational Interviewing (MI) techniques to modify behavior; and (iii) referral to treatment for those identified as needing more extensive treatment with access to specialist care. The roll out of the program involved a major effort to train midwives in screening with AUDIT and in motivating behavioral change using MI-techniques; a training likely to have enhanced the midwives’ ability to encourage health promoting behaviors also in domains other than alcohol. By studying heterogeneities—by type of medication and diagnosis, by age and socioeconomic status of mothers, and by sex of the child as well as the impact on the sex-ratio at birth—as well as maternal smoking and breast feeding, our aim is to provide insights into the mechanisms through which screening and BI for alcohol in antenatal care can affect child health. Interest in the effectiveness of universal alcohol prevention programs as an integral part of antenatal care, is motivated by a growing literature of well identified studies establishing a causal link between alcohol exposure in utero and negative birth outcomes (Wüst 2010 and Zhang 2010), school outcomes, educational attainment, labor market outcomes and a lower ratio of boys to girls (Nilsson, 2016) in observational data. While the negative effect of excess alcohol exposure, and binge drinking, has been widely accepted, the recent evidence puts a focus on likely negative effects also of low and moderate consumption (von Hinke et al. 2014). This recent evidence questions a large number of observational correlation studies suggesting that the risks of moderate consumption are ambiguous and depend on the nature of alcohol consumption (see meta studies by Polygenis et al. 1998; Abel and Hannigan, 1995). 7 The literature also refers to this type of public health program as SBIRT: Screening, Brief Intervention and Referral to Treatment, see eg Young (2014) et al for a review. Work in progress -­ do not quote IFAU Sober mom, healthy baby? 5 Interest in the effectiveness of this screening and BI program in antenatal care is also motivated by the large body of research on BI using MI. Such interventions are common and claimed to be effective in a number of areas of health: diabetes care, weight loss, smoking session, drug or alcohol addiction and in promoting reductions in risky behaviours (Rubak et al, 2004). However, in reviewing a large number of reviews, O’Donnell et al (2014) conclude that the evidence regarding interventions during pregnancy is yet rather weak. Moreover, studies of large scale BI-programs in primary care for general populations are rare and so is the evidence on effects of alcohol prevention on child health. To our knowledge this is the first attempt to evaluate the effects of a population wide nationally implemented screening and BI-program in maternity care on child health and maternal behavior. Due to timing constraints, not all antenatal clinics were able to introduce the program simultaneously (Socialstyrelsen, 2008). This resulted in a staggered introduction of the screening and MI across antenatal clinics in Sweden so that similar mothers giving birth in the years 2003-2009 faced different screening and alcohol prevention regimes depending on where they lived and when they were pregnant. This allows us to estimate the effects of the program with a difference-in-differences strategy. We use rich administrative data on prescription drugs and hospital care consumption (including detailed information on chemical classification and diagnosis) to construct measures of health, for the universe of first born children in Sweden during the implementation 2004-2009. In an additional analysis we use a similar strategy to estimate the effects on self-reported maternal behaviors and child health exploring survey data collected by the midwifes covering 70 percent of births during the years 2003-2008. We find that the program improves infant health, both as measured by pharmaceutical drugs and by inpatient care utilization during the first year of life. We also find evidence of reduced maternal smoking during pregnancy, and suggestive evidence of increased breastfeeding. In particular, we find that screening lowered the probability of children being prescribed a pharmaceutical drug during their first year of 8 A similar conclusion is drawn regarding other types of informational interventions to increase awareness of the risks of alcohol during pregnancy using various forms of media such as commercials, pamphlets etc (Crawford Williams et al 2015). 9 Nilsen et al 2012 analyze maternal self-reported (but anonymous) drinking habits pre-pregnancy and during pregnancy for mothers registered in antenatal care before and after the program was implemented in the municipality of Linköping. They find no significant differences in reported drinking habits but they do find improved perceptions of and a more positive attitude to the alcohol information received from the midwife. Work in progress -­ do not quote 6 IFAU Sober mom, healthy baby? life by 8.4 percent, and lowered the probability of being admitted to hospital during their first year of life with 7.5 percent. We find that the health effects are mainly driven by reductions in prescriptions related to infections and by reductions in inpatient care due to injury and ‘avoidable’ conditions, which would not have required hospitalization if the child had access to timely and effective preventive or primary care (e.g. asthma, diarrhea and infections). We find no effects on conditions that could be connected to congenital malformations or perinatal condition and complications at birth that would be associated with heavy alcohol exposure in early gestation. Neither do we find an effect on the sex ratio at birth nor do we find differential health effects by sex of the child. This pattern of results is consistent with the program having no influence on hazardous alcohol consumption in early gestation, which is what to expect given that it is administered towards the end of the first trimester. Instead, the results are consistent with the interpretation that the screening and brief alcohol intervention reduced alcohol exposure later in the pregnancy, leading to improvement in children’s immune system. The effects on avoidable conditions and injuries, as well as effects on maternal smoking cessation also point to behavioral effects that extend beyond alcohol consumption and the duration of the pregnancy. Effects on smoking may partly be the result of the MItraining improving midwives general ability to support health promoting behaviors, not only behaviors related to alcohol. Smoking and alcohol consumption are however often complements as is found in Wüst (2010). This paper is a contribution to the literature on the importance of in utero and early life conditions for child health by illustrating the importance of alcohol exposure and maternal behavior for child health. More specifically it is a contribution to the understanding for how policy interventions can impact child development. Our paper thus also contributes to the literature estimating effects of BI in general, and brief alcohol inventions in antenatal care in particular. Showing that the screening and BIprogram in Swedish antenatal care improved child health and maternal behaviors when implemented within the context of universally available antenatal care is an important argument for supporting such policy initiatives. The socioeconomic profile of the results also suggests that alcohol prevention in antenatal care contributes to closing socioeconomic gaps at birth. A further contribution of this paper is to the wider Work in progress -­ do not quote IFAU Sober mom, healthy baby? 7 literatures on screening and information interventions, and alcohol prevention in particular (O’Donnell et al, 2013). The rest of the paper is organized as follows. The following section reviews the literature on prenatal health and alcohol exposure. Section 3 summarizes antenatal care policies in Sweden and discusses the new screening and brief intervention program. In Section 4, we describe the empirical strategy and Section 5 describes the data. Finally, Section 6 reports the results from the main analysis and Section 7 reports the results using survey data. Section 8 concludes. 2 Prenatal health and alcohol exposure A large body of research documents the detrimental effects of severe alcohol exposure in utero (Abel, 1984, Streissguth et al., 1994). The most severe diagnosis associated with fetal alcohol exposure is Fetal alcohol syndrome (FAS) which includes a combination of congenital anomalies combined with confirmed maternal alcohol consumption during pregnancy, with the main symptoms being growth deficiency (both preand postnatal), FAS-specific facial features, and central nervous system damage causing cognitive and functional disabilities. Fetal alcohol spectrum disorders (FASD) is a non-diagnostic term for permanent birth defects (Sokol, Delaney-Black and Nordstrom, 2003), and includes a broader spectrum of growth deficiency and cognitive and psychosocial impairments and disabilities caused by the mother's consumption of alcohol during pregnancy (Streissguth et al. 1996; Clarke and Gibbard, 2003; Riley and McGee, 2005). While effects on the physical development of organs and extremities may be more affected at the early stages of gestation, there are reasons to believe that the development of the central nervous system and the brain as well as fetal growth and birth weight are sensitive to alcohol exposure throughout the pregnancy (eg Guerri, 2002). Although the link between heavy alcohol exposure and FAS is widely accepted, there are surprisingly few studies that can convincingly identify a causal relationship between alcohol consumption and child health in a general population of mothers. There are, however, a growing number of studies with well-identified causal effects utilizing sales restrictions to document the detrimental effects of maternal alcohol consumption on 10 See discussion in Nilsson (2015) for a discussion of the earlier mainly observational studies. Work in progress -­ do not quote 8 IFAU Sober mom, healthy baby? child outcomes at the population level (Zhang 2010, Fertig and Watson 2010, and Nilsson 2015). Zhang (2010) examines the relationship between drinking during pregnancy and infant birth outcomes utilizing changes in state-wide alcohol taxation. She finds that higher alcohol taxes reduce binge drinking among pregnant mothers and improves birth outcomes of children. This result is partly due to selection into motherhood, as unplanned pregnancies are more likely for women engaging in binge drinking. Similarly, Fertig and Watson (2010) find that changes in state minimum drinking age laws in the US have effects on infant health mainly by affecting the composition of families: alcohol availability by young adults is associated with more unplanned pregnancies, in particular among low SES parents. Composition effects are also found by Nilsson (2015) who studies a temporary (8.5 month) policy experiment of less restrictive sales rules for strong beer in two Swedish regions in the 1960’s. The experiment increased the availability of alcoholic beer for youths in the age 18-21 which increased alcohol consumption, most likely in the form of binge drinking. Nilsson also finds detrimental long run effects from alcohol exposure in utero in terms of substantially lower earnings, wages, educational attainments, and cognitive and noncognitive ability. The negative effects on earnings are found throughout the distribution but are largest below the median. The detrimental effects of increased alcohol availability are found to be strongest for fetuses exposed at early stages of the pregnancy, resulting in a higher than normal ratio of boys to girls and worse outcomes (educational attainment and earnings) for boys. 13 These studies suggest that maternal alcohol consumption, in particular the alcohol consumption of young mothers, is influenced by increased access to alcohol and that this increased consumption is harmful for children. von Hinke et al (2014) instead use so called Mendelinan randomization as a source of exogenous variation to identify effects of fetal alcohol exposure on the educational attainment of UK children. 11 Barreca and Page (2013) are however unable to find a significant effect. 12 The health of unplanned children is often worse since these children are more often born to lower SES mothers. 13 Effects on the sex-ratio, implying a lower ratio of boys to girls, are typically associated with negative shocks or presence of maternal stressors at the time of conception or during the first half of the pregnancy (Valente 2015). This effect is driven by selection at conception but also by spontaneous abortions and can be the result of different mechanisms with different implications for the sex difference in health of the children, conditional on live birth. Almond and Currie, 2011 find evidence of scarring, i.e. that differential survival would be the result of deteriorating maternal health during pregnancy resulting in a low sex-ratio and a sex gap in health at birth to the favour of girls. This is consistent with the findings of Nilsson 2015. Catalano et al 2008, however find evidence of so called culling, i.e. that the survival threshold of boys has shifted to the right such that surviving boys are in fact in better health. Work in progress -­ do not quote IFAU Sober mom, healthy baby? 9 Information on maternal genotypes of a particular gene, shown to influence alcohol metabolism and consumption, is used to instrument for alcohol use during pregnancy. Because carrying this variant of the gene affects alcohol consumption across individuals in the full population, they are able to study effects of low or moderate consumption in a representative population of mothers. The interesting feature with this study is that it shows that selection is the reason why OLS results indicate positive effects of wine consumption and moderate drinking throughout the pregnancy and negative effects of beer consumption and binge drinking. IV-estimates, instead are consistently negative suggesting that alcohol exposure is negative for educational attainment and that more alcohol, more binge drinking and longer exposure during the pregnancy is worse. Because the gene variant is likely to affect maternal alcohol consumption also after birth, it cannot be ruled out that both in utero and childhood exposure to maternal alcohol consumption matter for child outcomes. In a study on Danish register data, Wüst (2010) instead uses a sibling fixed effect approach to study the effects of alcohol consumption on child outcomes. She finds that controlling for selection using siblings turns the insignificant association between alcohol consumption and birth outcomes into a significant negative effect. As in the study of UK mothers, this reflects that mothers are positively selected into alcohol consumption during pregnancy. She also finds a dose–response relationship such that more drinking causes more harm, rather than finding that the effects are driven only by excessive consumption. 3 Antenatal Care, Screening and Brief Interventions Sweden has an extensive system of antenatal clinics, with an objective not only to strengthen parents in their parental role but also to detect and prevent poor health and offer support to mothers. The care received at the antenatal care clinics is free of charge and easily accessible. Health education is an important aspect of antenatal care and focuses mainly on lifestyle changes during pregnancy. Nearly 100 percent of all expecting mothers are enrolled in maternity care services delivered primarily through municipality-based public antenatal clinics (Socialstyrelsen, 2005); around 520 clinics in Sweden care for the about 100 000 pregnant women annually. During uncomplicated pregnancies, women typically have 6-10 prenatal visits to the antenatal clinic. The focus Work in progress -­ do not quote 10 IFAU Sober mom, healthy baby? of the first visit, which occurs around week 8-12 of the pregnancy, is primarily to make a physiological assessment and to provide information about pregnancy. An important aspect of health care during pregnancy is to identify risks and conditions—both medical and psychosocial—which can affect the pregnancy, the delivery, and the development of the fetus. By covering nearly all pregnant women in Sweden, the antenatal clinics have a strategic position in detecting and preventing prenatal alcohol exposure, and to provide support to women who experience difficulties to stop drinking alcohol during pregnancy. In 2004 the Risk Drinking project was initiated in Swedish maternity care in response of a growing concern for changed alcohol consumption patterns following Sweden’s entry to the EU. In particular, the alcohol consumption among women aged 28-38 increased during the late 1990’s (Bergman and Källmén, 2003). Since consumption of alcohol during pregnancy is influenced by established habits, changed consumption patterns in general, may have consequences for women's attitudes towards alcohol during pregnancy (Göransson, 2004). The Risk Drinking project was a nationwide effort to implement a brief alcohol intervention as an integral part of routine care. The project was run and financed by the Swedish Public Health Agency and had a large impact on the antenatal clinics’ alcohol preventive work by promoting the use of the AUDIT instrument to detect risky alcohol consumption (Socialstyrelsen, 2009); by introducing and providing training in MI as a tool for motivating reduced alcohol consumption; and by extra councelling and referral to specialists for mothers displaying a risky alcohol consumption pattern. AUDIT is a 10-item questionnaire, developed by WHO, covering three areas: consumption, addiction, and alcohol related damages (Babor et al., 2001). The AUDIT instrument was adapted for use in antenatal clinics by asking, not about present but rather, about pre-pregnancy alcohol behavior, and was promoted as a pedagogic tool to be used at the woman’s first visit at the antenatal clinic around week 8-12 of the pregnancy. The AUDIT questionnaire is filled out by the midwife or by the mother and is used as a basis for talking about alcohol habits. During the interview the midwife informs about risks with alcohol during pregnancy with the explicit purpose of motivating behavioral change among those who display risky consumption patterns. 14 MI is developed in Miller 1983 and Miller and Rollnick, 1991 15 See Appendix B for the AUDIT questionnaire. Work in progress -­ do not quote IFAU Sober mom, healthy baby? 11 This involves a motivational discussion exploring habits and the mother’s own positive and negative attitudes towards alcohol while maintaining an empathic, non-judgmental atmosphere. Based on the woman’s own ambivalence towards alcohol, the role of the midwife is to strengthen the woman’s own arguments against drinking by providing facts about the risks for the fetus. It is important that this is done in a compassionate way so as to avoid arguments and negative feelings that might evoke a defensive attitude. One strength of the AUDIT protocol is its sensitivity and high specificity— compared to other screening instruments—in detecting risky consumption at different levels of alcohol use and problems (Saunders et al., 1993, Reinert and Allen, 2007). Another strength lies in its implementation which is focused on women's alcohol consumption prior to pregnancy. Women are more likely to answer truthfully about pre-pregnancy consumption, and pre-pregnancy alcohol intake has been shown to be a good predictor of the alcohol consumption during pregnancy (Göransson et al., 2003). The AUDIT protocol grades alcohol behavior on a 0-40 scale, where a higher score indicates more hazardous alcohol consumption. Originally the cut-point for identifying at-risk drinking behavior in the general population to was set to 8. Studies later showed that the cut-point for women should be set lower and values of 5-6 or even as low as 3 has been suggested for identifying at-risk drinking among females (Reinert and Allen, 2007). If a woman scores a value of 6 or higher on AUDIT the midwife will immediately start a motivational BI with the aim of supporting modified behavior. The woman will also be invited for more frequent visits. If the midwife considers it necessary, or if the woman gets a very high AUDIT score, referral to other professions such as counselors, the social service, and/or an alcohol dependency clinic will also follow (Folkhälsoinstitutet 2014; Damström Thakker, 2011; Västra Götalandsregionen 2008). Importantly, the intervention is aimed at motivating and encouraging behavioral modification rather than coercion or merely providing health information. 16 See eg Handmaker and Wilborne (2001). 17 It is widely recognized that obtaining reliable self-reports of women's alcohol use during pregnancy is difficult because of stigma and because of uncertainty about what entails risky consumption (Gray and Henderson, 2006). 18 . Among those diagnosed as having hazardous or harmful alcohol use in a general population, 92% had an AUDIT score of 8 or more, and 94% of those with non-hazardous consumption had a score of less than 8 (Saunders et al., 1993). AUDIT scores in the range of 8-15 is found to represent a medium level of alcohol problems whereas scores of 16 and above represented a high level of alcohol problems. Since the effects of alcohol vary with average body weight and differences in metabolism, lowering the cut off for women with one point—i.e. to an AUDIT of 7—will increase sensitivity for this population groups (Babor et al., 2001). Work in progress -­ do not quote 12 IFAU Sober mom, healthy baby? During the roll out of the Risk Drinking project in antenatal care midwifes were trained in using AUDIT as well as in MI technique. The training programs were organized by the coordinating midwives at the county level. Training involved a full day training program on the risks of alcohol consumption during pregnancy and how to use the AUDIT questionnaire in antenatal care. A further important part of the program was training in MI techniques. This part of the program involved 3-4 days of training and recurring visits by instructors at the antenatal clinics in order to follow up and support implementation of AUDIT and MI. A limited number of lecturers and instructors were involved in these training programs and hence time constraints implied that it took some time to train midwifes in AUDIT and MI. As a result the program was gradually adopted by antenatal clinics, where the exact timing depended on accessibility and scheduling possibilities among both participating clinics and by lecturers and instructors. By 2010, 92 percent of the clinics had introduced AUDIT and MI (Socialstyrelsen, 2008). In an evaluation of the Risk Drinking project, the National Board of Public Health (Folkhälsoinstitutet, 2010) found that the fraction of midwives who thought they had good or very good knowledge about the risks of alcohol during pregnancy rose marginally between 2004 and 2009, from 94 to 99 percent. During the same period, the fraction midwives who judged their ability to identify at risk mothers as good or very good rose from 60 to 92 percent. In a survey of Stockholm midwives, midwives regarded MI-training, in particular, as very important in strengthening their ability to talk to mothers about alcohol (Damström Thakker, 2011). 4 Empirical strategy To estimate the effects of a universal screening brief alcohol intervention program in antenatal care on infant health and maternal behavior, we use a difference-in-differences approach where we utilize the staggered implementation of AUDIT screening and MI across antenatal clinics within counties. Although antenatal clinics are municipality 19 Together with Heads of Obstetrics, coordinating midwives in the counties are in charge of developing, implementing and evaluating local practice in the area of antenatal care and reproductive health. 20 In Figure A1 in Appendix A we describe the gradual implementation of the AUDIT-MI-program. 21 For a detailed account of the training program and implementation see eg Nilsen et al 2011. Details about the implementation are also based on an interview with Kerstin Petersson, head administrator of the MHV-register and Coordinating midwife in Stockholm County, October 16, 2015. Work in progress -­ do not quote IFAU Sober mom, healthy baby? 13 based, health care in Sweden is organized at the county level: 290 municipalities are divided into 21 counties which are responsible for the provision of health care. For this reason there is some regional variation in the organization and practices across different counties, which may affect health care utilization (Socialstyrelsen, 2011), and hence the measures of health used in this study. We will therefore focus on within-county variation between municipalities in the timing of implementation to identify the effects of the program. Figure 1 illustrates how the gradual increase in the share of antenatal clinics implementing the program yields a substantial municipal variation within counties (except for the counties of Uppsala, Jönköping, Gotland, Blekinge, and Västmanland) in the years before 2010. Figure 1. Regional implementation of the program by year Mothers are regarded as treated by the program if they—during the first four months of the pregnancy—live in a municipality where the antenatal clinics have implemented 22 33 municipalities are excluded from the analysis because the clinics within the municipality introduced the program in different years. The sample restrictions are discussed in Section 5.1. Work in progress -­ do not quote 14 IFAU Sober mom, healthy baby? the program, and the control group is pregnant women in other parts of the county where the program has not yet been introduced. The empirical model is given by: yy"#$% = αα + ββTTTTTTaaaaaaTTaaaa$% + γγ$ + ηη#% + θθbbaa" + XX"λλ + KK$%λλ + εε"#$%, (1) where yy"#$% is the outcome of child i in county c in municipality k, year t. With γγ$ being a vector of municipal fixed effects, and ηη#% a vector of county specific time effects, the variations between municipalities within a county identify the effect. TTTTTTaaaaaaTTaaaa$% is an indicator taking the value 1 if the mother belongs to a clinic which has implemented the screening and BI program and 0 otherwise. In order to control for seasonal patterns in infant health and drinking patterns we include an indicator for birth month, bbaa". XX" is a vector of controls for predetermined family characteristics. There is a social gradient both in child health (Cutler et al., 2008 and Mörk et al., 2014) as well as in drinking and awareness of the detrimental effects of alcohol consumption during pregnancy (Bergman and Källmén, 2003). We therefore include the following characteristics as controls: mothers’ and fathers’ age; immigrant status and educational level of the mother; whether the parents live together in the year that the child was born; and sex of the child. We also include municipal unemployment level and municipal alcohol sales per capita in the regression to control for time-varying differences in municipal characteristics, KK$%. The coefficient of interest is ββ, which is the estimate of the treatment effect. Standard errors are clustered at the municipal level. The main identifying assumption is that the timing of implementation is unrelated to changes in infant health and maternal alcohol consumption in the municipality. And since the timing of implementation was determined by when midwives could be scheduled for training in AUDIT and MI, rather than motivated by alcohol consumption patterns we believe that the parallel trends assumption is fulfilled. The assumption is corroborated by a number robustness tests in section 6.6. A potential threat to the identification comes from Swedish mothers being free to choose antenatal clinic. Mothers could potentially select into clinics based on their alcohol prevention practices: a woman with risky alcohol consumption could for example choose a clinic without screening if she is reluctant to reveal a potential abuse. In order to avoid this selection problem we restrict our attention to municipalities with Work in progress -­ do not quote IFAU Sober mom, healthy baby? 15 only one antenatal clinic or municipalities where all clinics implemented screening and BI at the same time. The problem of varying screening practices, and the scope for clinic choice, is more pronounced in larger cities with several clinics and in section 6.6 we present sensitivity analyses with regard to excluding these municipalities. Another potential threat to the identification strategy is that mothers who were exposed to the program at the antenatal clinic may also have been exposed to new alcohol preventive strategies elsewhere, e.g. at child health clinics after the child was born. Although not as well documented, the implementation of the Risk Drinking project in child health clinics was not coordinated with the implementation effort at antenatal clinics. In fact, child health clinics initiated the Risk Drinking project later and at a slower pace than the antenatal care clinics. In 2006, the fraction of child health nurses who had received at least some training in prevention of risky alcohol consumption was 52 percent, substantially lower than the corresponding fraction of midwives which was 88 percent. In addition, the midwives typically had received more training. By 2009, two thirds of midwives and one third child health nurses had received at least three days of training (Folkhälsomyndigheten, 2010) 4.1 Expected effects of the program In order to assess through which mechanisms a screening and brief alcohol intervention program for pregnant women affects infant health we analyze heterogeneities by different domains of infant health, by sex of the child and by socioeconomic status of the mother. The previous literature suggests that the type and timing of fetal alcohol exposure may give rise to different consequences. Exposure in early stages of gestation and heavy exposure through binging are likely to result in a skewed sex-ratio at birth (selectivity at conception and spontaneous abortion is more likely for boys) and potentially worse outcomes for boys (Valente, 2015). Long run, but moderate, exposure throughout the pregnancy, on the other hand, is more likely to have detrimental effects on the development of the central nervous system, the brain as well as fetal growth and birth weight (Guerri, 2002). 23 See Valente, 2015 for a thorough discussion of these mechanisms. Almond and Currie, 2011 find evidence of scarring, i.e. that differential survival would be the result of deteriorating maternal health during pregnancy resulting in a low boy-to-girl-ratio and a sex gap in health at birth to the favour of girls. This is consistent with the findings of Nilsson 2015. Catalano et al 2008, however find evidence of so called culling, i.e. that the survival threshold of boys has shifted to the right such that surviving boys are in fact in better health. Work in progress -­ do not quote 16 IFAU Sober mom, healthy baby? In order to capture effects of early and heavy alcohol exposure we specifically look at sex ratio at birth and gender heterogeneities in outcomes. Because the investigated screening and BI program takes place towards the end on the first trimester, we should not expect it to have any effects on alcohol exposure at the early stages of the pregnancy. Moreover, heavy abuse is likely to have been detected also before the introduction of the studied program. We therefore do not expect effects on sex ratios at birth or gender heterogeneities. To capture effects of fetal exposure throughout the pregnancy we instead study effects on infections which may be a consequence of increased sensitivity or reduced immune function related to birth weight and fetal growth (Gauthier, 2015). In addition, we study the most common diagnoses leading to hospitalization among infants, i.e. perinatal diagnoses, and respiratory conditions. Although these categories of diagnoses are more difficult to directly link to type of exposure they are more common among children with low birth weight. In order to capture post natal behavioral changes of the mother we look at injuries and a set of conditions which are considered as avoidable hospitalizations in the sense that appropriate care and nutrition are likely to reduce their incidence (Page et al. 2007). The program was designed to better detect at risk mothers. It is well known that the nature of alcohol consumption varies by maternal characteristics: younger and less educated women are more likely to engage in weekend binge drinking, whereas older and more educated women are more likely to have a consumption pattern with small or moderate quantities of alcohol on a more regular or every day basis (Wüst, 2010 and von Hinke Kessler Scholder, 2014). Differential effects by maternal age and education may thus pick up heterogeneous impact of the program due to heterogeneities in risk of alcohol exposed pregnancies as well heterogenous responses at given risk levels. Although the program was focused on alcohol prevention, it is possible that other behaviors are affected. We therefore also study effects on smoking and breastfeeding, 24 When using hospital admissions as outcome we combine we combine respiratory diagnoses (which include both admissions for asthmatic problems, croup, RS-virus and throat infections) and admissions for eye and ear infections. 25 These “avoidable” hospitalizations are admissions for certain acute illnesses and worsening chronic conditions that might not have required hospitalization if they had been managed through timely and effective utilization of primary care and through patient behavior. Note that all such hospitalizations cannot be avoided. Avoidable conditions fall into three categories: vaccine preventable, acute conditions, and chronic conditions; that, if managed well, should not require hospital admission. We use the definition for children suggested by the Public Health Information Development Unit in Australia (Page et al. 2007). Table A1 in Appendix A lists diagnoses groups and the ICD codes included as well as the ATC codes for the categories of drugs. Borttaget: l Work in progress -­ do not quote IFAU Sober mom, healthy baby? 17 which could be a consequence of reduced alcohol consumption, since alcohol and cigarettes are often consumed together, while mothers may be reluctant to breastfeed when they have been drinking. However, breastfeeding and smoking effects could also be spill-overs of MI training to other areas of health promotion if the midwives’ ability to successfully promote behavioral change is not limited to alcohol. 5 Data In the main analyses we combine data from administrative registers—e.g. the Population register, the Hospital Discharge register and the Prescription Drug register— with antenatal clinic level survey data on the implementation of the program from the Maternity Health Care Register. We describe these data below. In auxiliary analyses we also make use of individual level survey data from the Maternity Health Care Register. We describe these data in section 7 in connection to the results. 5.1 Study population and screening Our study population in the main analysis consists of all first-born children in Sweden born 2003-2009 and their parents. The population is identified through the population register held at Statistics Sweden. It covers all Swedish residents with information on year and month of birth, birth order and with a link to the biological parents. The analysis will focus only on first-time mothers since we want to avoid information given during earlier pregnancies to influence the results. Moreover, given the possibility that the program may affect the probability of having a second child, we avoid biases introduced by selection in second births by focusing on first borns. The sample is also restricted to include only children who are born in Sweden and whose mothers reside in Sweden, since we want to make sure that the mothers have been exposed to Swedish maternity care. For each parent we retrieve information on socioeconomic background characteristics from Statistics Sweden based on administrative records and population censuses; specifically: educational attainment, annual labor income, age, and municipality of residence. The information on educational attainment is based on a 3digit code, corresponding to the International Standard Classification of Education 1997. For earlier cohorts covered by this register, and for immigrants, information on educational attainment is obtained from census data, whereas the data for later cohorts Work in progress -­ do not quote 18 IFAU Sober mom, healthy baby? come directly from educational registers of high quality. The information on labor income stems from data that employers are mandated to report to the tax authorities for income tax declaration purposes. These data are matched with information on alcohol prevention practice at the municipal level using the municipality of residency of the mother. Data on the alcohol prevention at each antenatal clinic was collected by the Maternity Health Care Register. The register is managed by the medical profession and was initiated in 1999 in order to improve the quality and to enable monitoring and evaluation of the maternal health care. The register is based on a local organization of participating antenatal clinics. Participation by these facilities is not mandatory, yet in 2008 compliance was 89 percent. Since the register was initiated from within the profession and is used to benchmark quality and compare procedures, there is an incentive for accurate and high quality of reporting. Every year participating clinics submit information on working practices and services provided. We use this data to determine whether clinics are using a structured tool for alcohol screening for the period 2003-2008. Structured screening was first introduced as a part of the studied program and using structured screening implies that they have adapted the AUDIT instrument, MI-techniques and standardized procedures for referral to treatment. There is explicit information about the implementation of AUDIT screening from 2005 and onwards. For 2003 and 2004, clinics instead report whether they used “structured working methods to detect women with risky alcohol consumption”. For 2004 this implies AUDIT since the Risk Drinking project initiated the implementation of the program in 2004 and no alternative, structured screening methods were in use. Information on working methods at the antenatal clinics is linked to municipalities through the postal code. Most municipalities have only one antenatal clinic: Out of the 274 municipalities represented in Maternity Health Care Register, 72 municipalities have multiple clinics. Among municipalities with multiple units, 29 municipalities have units that introduced the screening simultaneously. Since we lack exact information on which center a woman visits we exclude the 33 municipalities where centers implemented the program in 26 For 2003 it is more ambiguous whether clinics responding that that use “structured working methods to detect women with risky alcohol consumption” in fact are using AUDIT, but it should be noted (i) that only 2 percent of the clinics were using such methods in 2003 as can be seen in Figure A1 in Appendix A, and (ii) that these clinics do not change screening status over the period. Details about the implementation are based on an interview with Kerstin Petersson, head administrator of the MHV-register and Coordinating midwife in Stockholm County, October 16, 2015. Work in progress -­ do not quote IFAU Sober mom, healthy baby? 19 different years. In total, pregnant women from 231 out of Sweden's 290 municipalities are included in the analysis. A mother is treated if she, when she was pregnant, lived in a municipality that had introduced structured screening. Since we have no information on the exact timing of the screening of women, we create a screening window consisting of the first four months of the pregnancy. Given that we do not have access to information about gestation weeks at birth, nor exact birth dates, we assume that all women are pregnant for 38 weeks, and that the child is born the first of each month. Since the first visit to the midwife usually occurs around week 8-12, screening is likely to fall within this four month window. To determine if a pregnant woman is affected by the program in a specific year, we restrict timing of treatment so that the full screening window has to occur past the turn of the year in order to belong to a "new" screening year. For example, a child born in August a given year is assumed to be conceived in November. Although the screening window overlaps the turn of the year, the treatment status of this child is determined by the screening regime the year prior to birth. In practice, this implies that children born between October and December in a given year are treated according to the screening practice in the birth year, whereas children born between January and September are treated according to screening practice the year prior to the birth year. The reason for the restrictive definition is that it is unlikely that all clinics implement the program in January but rather some time later during the year. Therefore, we also exclude the year of introduction in the main specification of the analysis. 5.2 Child health outcomes Our measures of child health are based on whether the child was admitted to hospital or was prescribed pharmaceutical drugs during the first (second) year of life. We create indicators for child health taking the value 1 if the child was admitted (over night) to hospital, respectively prescribed any drug, and 0 otherwise. Register information on all inpatient hospital episodes and on all prescribed pharmaceutical drugs purchased at pharmacies is available from the Swedish National Board for Health and Welfare. The hospital data includes detailed information on admission date and on primary and secondary diagnoses classified according to WHO’s ICD10 classification system. Hospitals are obliged by law to report this data, and the information is typically entered Work in progress -­ do not quote 20 IFAU Sober mom, healthy baby? into the hospital administrative system at discharge. Similarly, the drug data includes detailed information date of prescriptions and the chemical classification of the drug according to WHO’s ATC system. Pharmacies have strong incentives to report sales in order to get reimbursed from the public drug benefit. By using information from the ICD and ATC classification we define hospitalizations and drug prescriptions for different conditions and events of ill-health as described in Section 4.1 (see Table A1 in Appendix A for exact ICD10 and ATC codes). Information from the Hospital Discharge register is available for the whole implementation period 2003-2009. Information on drug prescriptions is available only from 2005-2009. 5.3 Descriptive statistics The first column of Table 1 displays summary statistics for the full population of firstborn children during the period 2003-2009. As discussed above we restrict the sample due to i) uncertainty of the exact month the screening was implemented, ii) uncertainty of exposure to screening in municipalities where some centers screened and others did not and iii) access to information on drug prescriptions. The second column includes information on the sample used in the analysis when studying hospitalization and the last column displays information on the sample when studying drug prescriptions. As can be seen from the first column, 17.3 percent of all first-borns during the period 20032009 are admitted to hospital during their first year of life. In our studied population the incidence is somewhat higher suggesting that hospitalization is more common in the included municipalities. Comparing column 1 to columns 2 and 3 also shows that there are some differences in the characteristics of the population. The reason is that municipalities which are excluded due to multiple antenatal clinics with different screening practices are larger cities with a higher share of single mothers, mothers with a higher education and a larger share of immigrant mothers. As can be seen in the last column, hospitalization is much less common than getting a drug prescribed during the first year of life, 18.7 percent of the children are admitted to hospital and 51.2 percent of the children get a drug prescribed. Over time the hospitalization rate of children has decreases somewhat whereas the share of children 27 The drug data only includes prescription drugs sold at pharmacies. Pharmaceutical drugs administered at hospitals or at primary care facilities are not covered. Work in progress -­ do not quote IFAU Sober mom, healthy baby? 21 getting drugs prescribed has been rather constant over the period (see Figure A2 and Figure A3 in Appendix A). It is worth noting that these two health measures may pick up different dimensions of health, in particular hospitalization reflects more severe or urgent health conditions. They may also pick up parental differences in health seeking behavior; if the parents refrain from seeking care in time the child may need hospital care for health problems which could have been resolved with a proper medication. Table 1. Sample characteristics Full population Hospital sample Drug sample (2003-­2009) (2003-­2009) (2005-­2009) Hospitalized children per 1000 173.1 188.9 187.3 (378.3) (391.4) (390.2) Children w drug prescript(%) 51.19 (49.99) Mother's age 29.02 28.29 28.27 (5.054) (5.043) (5.082) Father's age 31.96 31.41 31.42 (6.063) (6.150) (6.230) Single mother(%) 12.60 10.34 10.28 (33.18) (30.45) (30.38) University educ mother(%) 49.99 43.02 44.45 (50.00) (49.51) (49.69) Income below p20(%) 37.99 41.24 42.64 (48.54) (49.23) (49.46) Imigrant mother(%) 18.42 16.33 17.43 (38.77) (36.96) (37.93) Municipal unemployment(%) 3.514 3.545 3.385 (1.104) (1.185) (1.196) Observations 269819 108562 72690 5.4 AUDIT scores, maternal characteristics, behaviors and child outcomes Before proceeding to the analysis we characterize how maternal characteristics, health behaviors and child health relate to AUDIT scores. Table 2 presents statistics for first time mothers with AUDIT score 0-5; AUDIT score 6-9; with AUDIT score 10 and above. This description is based on individual level data from the Maternity Health Care Register for the period 2010-2014; that is, when the studied program is implemented throughout the country. We therefore have AUDIT scores for the vast majority of mothers. Table 2 reveals that for this later period, 9.6 percent of the pregnant women have elevated AUDIT scores of 6 or above at their sign in visit. Women with high AUDIT scores are younger than the average pregnant woman, and are more likely to have just Work in progress -­ do not quote 22 IFAU Sober mom, healthy baby? compulsory education. The fraction of non-Nordic immigrants with an elevated AUDIT score is lower than among women in general. About half of the first time pregnant women say they are in good or excellent health and 25 percent have normal BMI at registration. A remarkable difference between the different groups of women is that 24 percent of women with AUDIT ten or above smoked at registration while the corresponding fraction for low-AUDIT women was only 4 percent. This pattern also persists during pregnancy. Moreover, we see that fewer women with elevated AUDIT breastfeed fully or partially when the child is a month old. Table 2. Characteristics and behavior sign-­in visit AUDIT score 2010-­2014 AUDIT 0-­5 AUDIT 6-­9 AUDIT >= 10 characteristics of woman Age 29.1 27.7 26.3 young (<25) 0.21 0.32 0.47 old (>34) 0.16 0.10 0.08 university education 0.50 0.37 0.21 compulsory education 0.047 0.057 0.161 non-­nordic immigrant 0.15 0.04 0.04 in good health at registration 0.49 0.50 0.47 BMI normal at registration 24.3 24.5 24.5 smoking at registration 0.038 0.104 0.235 in good health during pregnancy 0.50 0.50 0.48 smoking in w 32 0.026 0.071 0.183 breastfeeding at 1 month 0.87 0.85 0.79 Observations 118496 11863 2256 6 Results We present the results of estimating the effect of implementing a screening and brief intervention alcohol prevention program in antenatal care on children's health. First we present results on the probability that the child is prescribed a drug or is admitted to hospital during the first years of life. Then we present results relating to specific health problems, heterogeneous effects across groups of mothers and whether screening pregnant women has differential effects on boys and girls or affects the sex ratio, and thereafter we analyze socioeconomic outcomes of parents. Finally we present some