Abstract
This study tested whether there is a linear or nonlinear relation between prenatal/birth cumulative risk and psychosocial outcomes during adolescence. Participants (n = 6,963) were taken from the Northern Finland Birth Cohort Study 1986. The majority of participants did not experience any contextual risk factors around the time of the child’s birth (58.1%). Even in this low-risk sample, cumulative contextual risk assessed around the time of children’s birth was related to seven different psychosocial outcomes 16 years later. There was some evidence for nonlinear effects, but only for substance-related outcomes; however, the form of the association depended on how the cumulative risk index was calculated. Gender did not moderate the relation between cumulative risk and any of the adolescent psychosocial outcomes. Results highlight the potential value of using the cumulative risk framework for identifying children at birth who are at risk for a range of poor psychosocial outcomes during adolescence.
Keywords: Cumulative Risk, Adolescence, Prenatal, Longitudinal, Substance, Academic
Multiple contextual factors have been found consistently to increase the likelihood that children and adolescents will experience poor adjustment outcomes (Cicchetti, 2016). These ecological factors include economic disadvantage, teenage pregnancy, poor educational attainment by parents, and parent substance misuse (for overviews, see Day, Ji, DuBois, Silverthorn, & Flay, 2016; January et al., 2017). Contextual risk factors typically do not occur in isolation and thus children who experience one risk factor often are exposed to others (Evans, Li, & Sepanski Whipple, 2013; Felitti et al., 1998; McLaughlin et al., 2012). To understand the influence of contextual risk on child and adolescent psychosocial adjustment, it is important to consider the co-occurrence of adverse life circumstances (McLaughlin & Sheridan, 2016).
A continued challenge for researchers, practitioners, and policy makers interested in children’s emotional and behavioral health is how best to conceptualize and quantify exposure to multiple contextual risk factors. One approach is the cumulative risk model (Rutter, 1979). This model posits that the number of risk factors rather than any specific one provides a useful metric for capturing children’s level of risk for developing psychosocial difficulties. Individual risk factors in this approach are typically dichotomized and then summed to create a cumulative risk index. A recent review highlights key contributions of the cumulative risk model and describes several advantages of this approach compared to other methods of accounting for children’s exposure to multiple risk factors including reduced measurement error, enhanced validity, and increased statistical power (Evans et al., 2013). This review also discussed several important unanswered questions. One fundamental issue that warrants further attention is determining whether the additive assumption underlying the cumulative risk model best represents the relation between cumulative risk and psychosocial problems.
Most cumulative risk research has examined a linear, additive model in which increases in the number of risk factors correspond to increases in the level of adjustment problems (Flouri, 2008). It is possible, however, that a threshold exists and the relation between the cumulative risk index and psychosocial outcomes changes after a certain number of risk factors. A nonlinear relation between cumulative risk and psychosocial outcomes indicates there is a threshold after which the relation changes. The change following a threshold could take one of two forms: (a) an accelerating effect in which lower levels of risk impact an outcome gradually up to a certain point (e.g., 4 risk factors) after which the negative impact on individuals intensifies or (b) a saturation effect in which the impact of cumulative risk on an outcome builds steadily to a plateau and levels off thereafter (Horan & Widom, 2015).
Evans et al. (2013) documented in their review whether each cumulative risk study provided support for a linear or nonlinear effect (see supplemental Table A in Evans et al., 2013). Although there are notable exceptions (e.g., Appleyard, Egeland, van Dulmen, & Sroufe, 2005; Horan & Widom, 2015), relatively few studies provided formal, direct tests of whether a linear or nonlinear relation best accounted for the cumulative risk-outcomes associations. For studies that did not conduct a direct test, Evans et al. used descriptive data to examine whether findings were consistent with a linear or nonlinear relation between cumulative risk and outcomes. They concluded that although caution should be taken when interpreting their observations “an equal proportion of CR studies found linear as opposed to nonlinear functions” (p. 1346). Thus, the evidence for a linear, additive compared to a nonlinear association between cumulative risk and psychosocial outcomes is mixed. Moreover, as discussed by Horan and Widom (2015), there is support for both the accelerating and saturation nonlinear effects. The current study was designed to help clarify the mixed results by directly testing linear and nonlinear relations between cumulative risk and psychosocial outcomes using a large representative sample with a longitudinal design spanning 16 years (see below for details).
Some cumulative risk studies have found linear and nonlinear effects for different outcomes within the same study (see supplemental Table A in Evans et al., 2013). For example, Horan and Widom (2015) found a quadratic relation between cumulative risk and years of education but a linear relation between cumulative risk and mental health problems and criminal arrests. Results like this highlight the need to not only directly examine linear and nonlinear effects but also consider multiple outcomes within the same study. Substance-related problems, delinquency, emotional difficulties, and academic problems are common during adolescence (Merikangas et al., 2010) and are related to a wide range of poor psychosocial outcomes later in life (Copeland et al., 2013). There continues to be a need to better understand risk factors that contribute to psychosocial difficulties among adolescents.
Another way to build on prior research examining the relation between cumulative risk and the psychosocial adjustment of youth is to identify factors that moderate the association. The review by Evans et al. (2013) highlighted several studies that found youth gender to be a moderator of the relation between cumulative risk and youth adjustment. Moreover, recent studies suggest there is a nuanced role for gender as a moderator of that relation. For example, Buehler and Gerard (2013) found that cumulative family risk was related to internalizing problems only for girls; whereas, the cumulative family risk index was related to externalizing problems only for boys. Horan and Widom (2015) found that gender moderated the relation between cumulative risk and educational attainment and adult criminal arrests but not anxiety and depressive symptoms. Wong et al. (2013) found a linear relation between cumulative risk and delinquency for boys and a quadratic relation for girls. Our own prior work using the same sample as the present study found that cumulative contextual risk at the prenatal/birth period was a significant positive predictor of adolescent risky sexual behavior for girls but not boys (Mason et al., 2016). Additional research is needed to better understand the seemingly complex role of gender in moderating the cumulative risk-youth outcomes relations.
A limitation of the cumulative risk literature is that risk factors are often assessed retrospectively and aggregated across developmental periods. The impact of cumulative risk on psychosocial adjustment can be better understood by considering risk at specific developmental periods. Prenatal and early childhood risk factors have been shown to have a particularly strong short- and long-term impact on emotional and behavioral outcomes (Shonkoff et al., 2012). Relatively few cumulative risk studies, however, have focused exclusively on prenatal risk factors (Evans et al., 2013). An advantage of investigating cumulative risk during the prenatal period is that associations with child and adolescent outcomes are not confounded with the context of risk. Thus, stronger conclusions can be made regarding the association between cumulative risk and outcomes.
Present Study
The present study investigated whether there is a linear or nonlinear relation between prenatal cumulative contextual risk and psychosocial outcomes during adolescence. The study extends prior research in several important ways. First, formal, direct tests of linear and nonlinear effects of cumulative risk on adjustment difficulties were conducted. Second, contextual risk factors were assessed during the prenatal period. Third, multiple psychosocial outcomes during adolescence (16 years after the assessment of risk) were considered in the same study. The psychosocial outcomes included perceived academic performance, internalizing symptoms, externalizing symptoms, substance misuse (cigarette use, heavy episodic drinking, and illicit drug use), and criminal behavior. This study also examined whether gender moderated the relation between prenatal cumulative contextual risk and adolescent psychosocial outcomes.
Data from a large birth cohort (Northern Finland Birth Cohort 1986 Study [NFBC1986]) was used. Several aspects of the health care system in Finland make the NFBC1986 well-suited to examine prenatal/birth risk factors and later psychosocial outcomes. Finland has an extensive network of local maternity health care centers governed by municipal districts. They are free of charge and most expecting mothers attend them monthly during pregnancy as well as following pregnancy for at least the first six months. Thus, most pregnancies and infants are carefully monitored, which facilitates early detection and treatment of any complications or risks. Mothers also get additional support in the form of a “maternity package” with abundant clothing and other materials for taking care of newborns. Parental leave is covered as a universal welfare provision and lasts about one year. Either the mother or father is eligible for this benefit with additional leave available for the other parent, typically the father. Finnish children have access to free municipal day care (or equivalent) during the first 3 years, after which access is heavily subsidized by the government. In general, local health centers provide services and treatments free of charge for all children and adults. The considerable health-related resources available to children and families in Finland have the potential to influence the relations between prenatal risk factors and later psychosocial outcomes. The NFBC1986 data thus provide a unique opportunity to directly test linear and nonlinear cumulative risk effects.
Method
Participants and Procedures
Participants were taken from the NFBC1986, a general population-based study of individuals with expected birthdates between July 1, 1985 and June 30, 1986 from the two northernmost provinces of Finland (Oulu and Lapland). There were 9,479 births, representing 99.0% of all deliveries in the provinces during that timeframe. Additional details regarding the data collection procedures are provided elsewhere (Hurtig et al., 2007; Jarvelin, Hartikainen, & Rantakallio, 1993). The study was approved by the ethical committee of the Northern Ostrobothnia Hospital District.
The current analysis sample included adolescents with some self-report data. Specifically, 92% (8,755 of 9,479) of parents provided consent and youth provided assent for data collected. Eighty percent (7,039 of 8,755) of those consented and assented had some adolescent self-report data. In the 7,039 consented cases, there were 74 sets of twins and 1 set of triplets. For each set of non-singletons, one child was randomly selected for the analysis dataset (thus dropping 76 children), yielding a final analysis dataset of 6,963 participants (73% of original birth cohort; Miettunen et al., 2014). Forty-nine percent of participants in the analysis sample were male and the mean age of adolescent participants at the time of data collection was 16.0 with ages ranging from 14.58 to 16.96 years.
Attrition analyses were conducted to compare the Birth Cohort (N = 9432) to the Analysis Sample for the present study (N = 6963). Fourteen prenatal/birth variables were compared. Four statistically significant differences were found (28.6%). Specifically, there were more girls in the Analysis Sample (51.0%) compared to the Birth Cohort (48.4%); there were more children born below 2,500 grams in the Birth Cohort (3.7%) compared to the Analysis Sample (3.0%); there were more single parent families in the Birth Cohort (5.1%) compared to the Analysis Sample (4.4%); and there were more mothers who reported smoking while pregnant in the Birth Cohort (15.9%) compared to the Analysis Sample (14.4%). These findings provide some evidence that the Analysis Sample was at somewhat lower risk than the Birth Cohort.
Data collected during three developmental periods were utilized for the present study: (a) prenatal/birth, (b) middle adolescence, and (c) late adolescence. Prenatal/Birth: At their first antenatal visit to the local prenatal clinic, mothers were given a prenatal background questionnaire, with instructions to return by their 24th gestational week. Topics of this questionnaire included the mother’s health and pregnancy history, as well as both mother and father substance use, education and employment. Medical data on pregnancy and delivery was completed by midwives and/or medical staff at the prenatal clinics. Middle Adolescence: In 2000 – 2001, participants (aged 15–16) were asked to complete a postal questionnaire on health, living habits, and social background. Eighty percent (7344 of 9215) completed this questionnaire. They were also invited to a clinical examination, with a participation rate of 76% (6985 of 9215). During this examination, participants completed a questionnaire about their eating habits, stress, sexual behavior, substance use, and mental well-being (Hurtig et al., 2007). Late Adolescence: Court criminal records were obtained for the birth cohort during late adolescence, spanning the years of 2003 through 2005.
Measures
Cumulative Contextual Risk during the prenatal/birth period was measured with ten indicators. All measures were collected through the pregnancy questionnaire completed by mothers, with the exception of birth weight (provided by medical staff at time of delivery). Ten indicators were selected: low birth weight, teenage motherhood, single parent status, multiple unions, maternal school dropout, smoking while pregnant, drinking while pregnant, paternal drinking, economic exclusion, and material deprivation. These indicators were chosen because they are associated with a wide range of psychosocial outcomes during adolescence and have been used in prior cumulative risk research. A recent study using the NFBC1986 data provided a rationale (including key references) for including each indicator in a cumulative risk index (for details, see January et al., 2017). A description of each indicator is provided below. Indicators were coded as 1 to represent presence of risk (as defined below) and as 0 to represent absence of risk. The distribution of the cumulative risk index appears in Table 1. As shown, although the possible range for the index was 0 to 10, the index actually ranged from 0 to 5.
Table 1.
Correspondence between Number of Risk Factors and Psychosocial Outcomes
| Number of Risk Factors | N | (%) | Perceived Academic Difficulties N = 6513 M (SD) | Internalizing Symptoms N = 6488 M (SD) | Externalizing Symptoms N = 6502 M (SD) | Cigarette Use N = 6443 M (SD) | Heavy Drinking N = 6183 M (SD) | Illicit Drug Use N = 6090 % Endorsed | Criminal Behavior N = 6952 % Endorsed |
|---|---|---|---|---|---|---|---|---|---|
| Number of Risk Factors using CR Index Ranging from 0 to 5 | |||||||||
| 0 | 4048 | (58.1%) | 3.38 (2.03) | 39.75 (7.02) | 38.47 (6.24) | 0.85 (1.62) | 0.63 (1.02) | 4.9 | 15.4 |
| 1 | 1859 | (26.7%) | 3.71 (2.03) | 40.27 (7.33) | 39.28 (6.39) | 1.14 (1.85) | 0.82 (1.11) | 6.9 | 16.9 |
| 2 | 696 | (10.0%) | 3.85 (1.98) | 40.61 (7.48) | 40.42 (7.26) | 1.52 (2.10) | 0.95 (1.17) | 9.4 | 22.1 |
| 3 | 237 | (3.4%) | 4.28 (2.20) | 41.20 (8.03) | 41.05 (6.62) | 1.78 (2.22) | 1.15 (1.32) | 9.2 | 19.4 |
| 4 | 91 | (1.3%) | 4.77 (2.07) | 40.00 (7.05) | 41.73 (7.05) | 2.36 (2.29) | 1.06 (1.15) | 11.4 | 18.7 |
| 5 | 32 | (0.5%) | 4.07 (2.11) | 40.00 (7.47) | 39.54 (5.87) | 2.15 (2.48) | 0.89 (1.15) | 7.4 | 15.6 |
| Number of Risk Factors Using CR Index Ranging from 0 to 4a | |||||||||
| 4 or more | 123 | (1.8%) | 4.60 (2.10) | 40.00 (7.12) | 41.21 (6.83) | 2.31 (2.33) | 1.02 (1.15) | 10.4 | 17.9 |
| Number of Risk Factors Using CR Index Ranging from 0 to 3b | |||||||||
| 3 or more | 360 | (5.2%) | 4.39 (2.17) | 40.80 (7.75) | 41.10 (6.68) | 1.96 (2.27) | 1.10 (1.26) | 9.6 | 18.9 |
|
| |||||||||
| Entire Sample | 6963 | (100%) | 3.56 (2.05) | 40.03 (7.19) | 39.01 (6.45) | 1.05 (1.80) | 0.74 (1.08) | 6.1 | 16.6 |
Note. N for Cumulative Risk Index = 6963. CR = Cumulative Risk.
Values for 0, 1, 2, and 3 risk factors are the same as those reported for the CR Index ranging from 0 to 5.
Values for 0, 1, and 2 risk factors are the same as those reported for the CR Index ranging from 0 to 5.
Adolescents were asked to report about their most recent grades for the measure of Perceived Academic Difficulties. The timeframe for reporting Internalizing and Externalizing Symptoms was prior six months. Timeframes for Cigarette Use, Heaving Drinking, and Illicit Drug Use were current, past 30 days, and lifetime, respectively. Information about Criminal Behavior was collected using official court data on criminal sanctions when adolescents were between 16.5 and 20.7 years old.
Low birth weight was coded 1 if the child was born weighing under 2,500 grams (Zegers-Hochschile et al., 2009). Teenage motherhood was coded 1 if the mother was 19 or younger when she gave birth to the participant. Single parent status was coded 1 if the mother was not married or cohabitating (i.e., sharing a household with a romantic or other partner). Multiple unions was coded 1 if the mother had at least one prior registered union, such as marriage or cohabitation (note that this coding assigns zero to two kinds of mothers: those in a first relationship and those with no prior or current relationship). Maternal school dropout was coded 1 if the mother had completed fewer than 9 years of comprehensive schooling (Grades 1–9). Smoking while pregnant was coded 1 if the mother smoked after the first trimester of pregnancy. If the mother drank alcohol during pregnancy, drinking while pregnant was coded 1. Paternal drinking was coded 1 if the mother reported that the child’s father had five or more alcoholic drinks per typical week. Economic exclusion serves as the household’s socioeconomic status, and was coded 1 if the highest occupational status of the household adults was either unskilled worker (manual labor), unemployed, or on disability pension. Material deprivation was coded 1 if the household had one or none of the following four items: washing machine, telephone, flushing toilet, or indoor bathroom.
Perceived Academic Difficulties was the adolescent’s mean self-rating of competence in four core subjects: science, math, humanities, and Finnish (α = .76). Each of the four survey items began with “Compared to other pupils of your age, how well are you doing in the following school subjects” with response categories of 4 = “better than average”, 3 = “average”, 2 = “worse than average”, and 1 = “really badly”. The four items were recoded so that higher scores corresponded to more difficulties.
Internalizing Symptoms were obtained from the Youth Self-Report (YSR; Achenbach, 1991) collected via postal questionnaire during adolescence. The 21-item internalizing scale was used in this study. For each item, youth selected how true that item was for them in the past 6 months, using response options of: 0 “not true”, 1 “somewhat or sometimes true”, or 2 “very true or often true.” The suicidality item, “I think about killing myself,” was modified for the purposes of this study to “I think seriously about harming myself”. Raw scores were summed per Achenbach scoring algorithm (Achenbach, 1991) and only computed if 67% of scale items had valid responses (M = 40.03, SD = 7.19, range = 23 to 81). In the current sample, the internal consistency of the Internalizing Symptoms scale was good (α = .88).
Externalizing Symptoms were assessed with 19 of 20 items from the externalizing scale of the YSR (Achenbach, 1991). The substance item, “I use alcohol or other drugs for nonmedical purposes” was excluded from the scale to avoid overlap with the adolescent substance misuse items. For each item, youth selected how true that item was for them in the past 6 months, using response options of: 0 = “not true”, 1 = “somewhat or sometimes true”, or 2 = “very true or often true.” Raw scores were summed per Achenbach scoring algorithm (Achenbach, 1991) and only computed if 67% of scale items had valid responses (M = 39.01, SD = 6.45, range = 25 to 76). In the current sample, the internal consistency of the Externalizing Symptoms scale was good (α = .86).
Adolescent Substance Misuse was represented by three variables: (a) current cigarette use, (b) heavy drinking, and (c) lifetime illicit drug use. Current cigarette use was determined by the question, “Do you smoke now?” with responses coded as 0 “not at all”, 1 “occasionally”, 2 “one day a week”, 3 “2–4 days a week”, 4 “5–6 days a week”, and 5 “7 days a week.” Heavy drinking was measured with a question on binge drinking, with variation in amount by gender: “Think back for the past 30 days. How many times during that time have you drunk <Boys: SIX DRINKS> <Girls: FOUR DRINKS> or more on the SAME OCCASION?” with responses coded as: 0 “never”, 1 “once”, 2 “twice”, 3 “3–5 times”, 4 “6–9 times”, and 5 “10 times or more”. Lifetime illicit drug use was measured with three questions about marijuana, hard drugs, or intravenous drug use. Each question used the stem, “Have you ever tried or used any of the following substances:”, followed by “marijuana or hashish?”, “Ecstasy, heroin, cocaine, amphetamine, LSD or other similar drugs?”, or “intravenously injected drugs?” Response options were never, once, 2–4 times, 5 times or more, or regular user. The prevalence rates for the three categories of illicit drugs were as follow: 6% used marijuana or hashish one or more times, .5% used ecstasy, heroin, cocaine, amphetamine, LSD or other similar drugs one or more times, and .1% intravenously injected drugs one or more times. These three items were collapsed into a single dichotomous lifetime illicit drug use (1) or non-use (0) variable due to low prevalence rates.
Criminal Behavior was assessed using official court data on criminal sanctions. These court data are based on events from January 1, 2003 through December 31, 2005, when the birth cohort members ranged from 16.5 to 20.7 years of age. In Finland, individuals cannot have a criminal record until age 15. The Criminal Behavior variable is a dichotomization of the count of sanctions disposed in criminal court.
A variable representing the Number of Missing Cumulative Risk Indicators was created to account for the fact that some participants did not have complete data on all of the cumulative risk indicators. As noted, the cumulative risk index was created by summing the scores on each of the 10 indicators. If a participant had a score on at least one indicator, he/she received a cumulative risk index score. The Number of Missing Cumulative Risk Indicators variable was the sum of the number of missing cumulative risk indicators (M = .86; SD = 1.57; range = 0 to 8). The following were frequencies (and percentages) for the Number of Missing variable: 0 = 4476 (64.3%), 1 = 1259 (18.1%), 2 = 344 (4.9%), 3 = 186 (2.7%), 4 = 221 (3.2%), 5 = 330 (4.7%), 6 = 95 (1.4%), 7 = 39 (0.6%), and 8 = 13 (0.2%). Thus, the majority of participants had missing data on 1 or fewer indicators.
Data Analysis
Path analyses were conducted to examine whether linear or nonlinear effects best accounted for the relations between prenatal/birth cumulative risk and psychosocial outcomes during adolescence. Two path models were conducted. In the first path model, independent variables were the cumulative risk index (linear effect) and the cumulative risk index squared (nonlinear effect). Gender and the variable representing the number of missing cumulative risk indicators were included as covariates. Dependent variables were measures of Perceived Academic Difficulties, Internalizing Symptoms, Externalizing Symptoms, Cigarette Use, Heavy Episodic Drinking, Illicit Drug Use, and Criminal Behavior. All seven dependent variables were examined simultaneously.
The second path model examined whether gender moderated the relations between prenatal/birth cumulative risk and psychosocial outcomes. Two product terms were added to the independent variables included in the prior model: cumulative risk index (linear effect) X gender and cumulative risk index squared (nonlinear effect) X gender. The same dependent variables were investigated.
Path models were estimated using Mplus Version 7.11 (Muthén & Muthén, 1998 – 2012). Several estimation methods are available in Mplus for conducting analyses with a combination of continuous and categorical dependent variables (Muthén, Muthén, & Asparouhav, 2015). Full Information Maximum Likelihood (FIML) estimation with robust standard errors was used because it offers advantages when working with datasets that contain missing data (Muthén et al., 2015). The entire sample was retained in the path analyses because all available data are used in the FIML estimation process. FIML has been shown to produce unbiased parameter estimates and standard errors (e.g., Hallgren & Witkiewitz, 2013). The path models estimated were just-identified and thus model fit could not be evaluated.
Results
As shown in Table 1, more than half of the participants did not experience any prenatal/birth risk factors (58.1%). Fifteen percent of participants experienced 2 or more prenatal risk factors and less than 2% experienced 4 or more. The average number of prenatal risk factors was .64 (SD = .92). The correspondence between number of risk factors (using the cumulative risk index ranging from 0 to 5) and each psychosocial outcome is also shown toward the top of Table 1. One general pattern was evident for all seven outcomes. Specifically, at lower levels of risk, the severity (or prevalence) of problems increased as the number of risk factors increased. At higher levels of risk, the severity of problems slightly decreased (see top of Table 1).
As noted, a path model was used to investigate whether linear or nonlinear effects best accounted for the relation between cumulative risk at birth (using cumulative risk index ranging from 0 to 5) and psychosocial outcomes during adolescence (see Table 2). Findings indicated that the linear cumulative risk index was a significant positive predictor of all seven outcomes. The nonlinear cumulative risk variable was found to be a significant negative predictor of Heavy Drinking and Illicit Drug Use. These negative quadratic terms are consistent with a saturation effect in which differences existed at lower levels of risk that plateaued at higher levels (see Table 1 for pattern of means/prevalences). The nonlinear cumulative risk variable was not significantly related to Perceived Academic Difficulties, Internalizing Symptoms, Externalizing Symptoms, Cigarette Use, or Criminal Behavior.
Table 2.
Regression Analyses with Linear and Nonlinear Cumulative Risk (Ranging from 0 to 5) Predicting Psychosocial Outcomes
| Predictors | Perceived Academic Difficulties β | Internalizing Symptoms β | Externalizing Symptoms β | Cigarette Use β | Heavy Drinking β | Illicit Drug Use OR | Criminal Behavior OR |
|---|---|---|---|---|---|---|---|
| Model 1 | |||||||
| Gendera | .00 | −.72*** | −.27*** | −.11*** | −.01 | .78* | 3.89*** |
| Number of Missing CR Variables | .05*** | .00 | .04** | .05*** | .05*** | 1.10** | 1.04 |
| CR Linear | .15*** | .08** | .17*** | .17*** | .22*** | 1.62*** | 1.33*** |
| CR Quadratic | −.03 | −.05 | −.06 | .00 | −.09** | .93* | .96 |
| Model 2 | |||||||
| Gendera | .02 | −.72*** | −.29*** | −.11*** | −.02 | .72* | 3.64*** |
| Number of Missing CR Variables | .05*** | .00 | .04** | .05*** | .06*** | 1.10** | 1.04 |
| CR Linear | .18*** | .08 | .13** | .18*** | .19*** | 1.44* | 1.16 |
| CR Quadratic | −.05 | −.05 | −.03 | −.01 | −.06 | .97 | .99 |
| CR Linear X Gender | −.05 | .00 | .06 | −.01 | .04 | 1.36 | 1.21 |
| CR Quadratic X Gender | .03 | .00 | −.04 | .02 | −.04 | .90 | .96 |
Note. N = 6963. CR = Cumulative Risk. β = standardized regression coefficients. OR = Odds Ratio. Girls = 0 and Boys = 1.
p < .05.
p < .01.
p < .001.
The CR Index that ranged from 0 to 5 was used in these analyses.
Because Gender is a binary independent variable, the STDY Standardization values are reported for all continuous outcome variables (Muthén & Muthén, 1998 – 2012).
A second path model was conducted to examine whether gender moderated the relation between cumulative risk and psychosocial outcomes. As shown in Table 2, no significant CR Linear X Gender or CR Quadratic X Gender relations were found.
As can be seen at the top of Table 1, the percentages of participants who experienced 3, 4, or 5 risk factors was relatively small (3.4%, 1.3% and 0.5%, respectively). The relatively small number of individuals at the highest levels of risk could have an undue influence on the relations between cumulative risk and adolescent psychosocial outcomes. To examine this possibility, we conducted the analyses two additional times. In one set of follow-up analyses, we truncated the number of risk factors at four (i.e., combined those who experienced 4 or 5 risk factors into a 4 or more category). In the other set of follow-up analyses, we truncated the number of risk factors at three (i.e., combined those who experienced 3, 4, or 5 risk factors into a 3 or more category). The follow-up analyses using the two truncated cumulative risk variables were conducted only because of the small number of individuals who experienced 3 or more risk factors. The two truncation points were selected based solely on the small percentages of individuals at the 3, 4, and 5 risk levels. The cumulative risk index was not truncated any further because a relatively large percentage of individuals were exposed to 2 risk factors (10%).
Table 1 provides descriptive information about the correspondence between the number of risk factors truncated at four and each psychosocial outcome (see section labeled Number of Risk Factors Using CR Index Ranging from 0 to 4). Two patterns emerged. For measures of Perceived Academic Problems, Externalizing Symptoms, Cigarette Use, and Illicit Drug Use, the severity of problems increased as the number of risk factors increased across the range of risk (no decrease in problems at high levels of risk was observed). For the measures of Internalizing Symptoms, Heavy Drinking, and Criminal Behavior, the severity (or prevalence) of problems increased as the number of risk factors increased at lower levels. At higher levels of risk, however, the severity of problems slightly decreased. Findings from the path model using the truncated at four risk variable are shown in Table 3. Similar to findings for the initial risk variable, the linear cumulative risk index was a significant positive predictor of all seven outcomes. Moreover, the nonlinear cumulative risk variable truncated at four was a significant negative predictor of Heavy Drinking. No significant CR Linear X Gender or CR Quadratic X Gender relations were found in analyses using the cumulative risk index truncated at four.
Table 3.
Regression Analyses with Linear and Nonlinear Cumulative Risk (Ranging from 0 to 4) Predicting Psychosocial Outcomes
| Predictors | Perceived Academic Difficulties β | Internalizing Symptoms β | Externalizing Symptoms β | Cigarette Use β | Heavy Drinking β | Illicit Drug Use OR | Criminal Behavior OR |
|---|---|---|---|---|---|---|---|
| Model 1 | |||||||
| Gendera | .00 | −.72*** | −.27*** | −.11*** | −.01 | .78* | 3.89*** |
| Number of Missing CR Variables | .05*** | .00 | .04** | .05*** | .06*** | 1.10** | 1.04 |
| CR Linear | .13*** | .09** | .15*** | .15*** | .21*** | 1.62*** | 1.32** |
| CR Quadratic | .00 | −.05 | −.04 | .02 | −.08* | .93 | .96 |
| Model 2 | |||||||
| Gendera | .02 | −.72*** | −.29*** | −.11*** | −.02 | .72* | 3.65*** |
| Number of Missing CR Variables | .05*** | .00 | .04** | .05*** | .05*** | 1.10** | 1.04 |
| CR Linear | .15*** | .09* | .11** | .15** | .18*** | 1.43* | 1.17 |
| CR Quadratic | −.02 | −.06 | −.01 | .01 | −.05 | .97 | .99 |
| CR Linear X Gender | −.04 | −.01 | .06 | .00 | .04 | 1.35 | 1.19 |
| CR Quadratic X Gender | .02 | .01 | −.04 | .01 | −.04 | .91 | .96 |
Note. N = 6963. CR = Cumulative Risk. β = standardized regression coefficients. OR = Odds Ratio. Girls = 0 and Boys = 1.
p < .05.
p < .01.
p < .001.
The CR Index that ranged from 0 to 4 was used in these analyses.
Because Gender is a binary independent variable, the STDY Standardization values are reported for all continuous outcome variables (Muthén & Muthén, 1998 – 2012).
Table 1 also provides descriptive information about the correspondence between the number of prenatal/birth risk factors truncated at three and each psychosocial outcome. Two patterns emerged. For all measures except Criminal Behavior, the severity of problems increased as the number of risk factors increased across the range of risk (no decrease in problems at high levels of risk was observed). For Criminal Behavior, the prevalence of problems increased as the number of risk factors increased at all levels except the highest. At the highest level of risk, the prevalence slightly decreased. Findings from the path model using the truncated at three risk variable are shown in Table 4. The linear cumulative risk index was a significant positive predictor of all seven outcomes. The nonlinear cumulative risk variable truncated at three was not a significant predictor of any of the psychosocial outcomes. No significant CR Linear X Gender or CR Quadratic X Gender relations were found in analyses using the cumulative risk index truncated at three.
Table 4.
Regression Analyses with Linear and Nonlinear Cumulative Risk (Ranging from 0 to 3) Predicting Psychosocial Outcomes
| Predictors | Perceived Academic Difficulties β | Internalizing Symptoms β | Externalizing Symptoms β | Cigarette Use β | Heavy Drinking β | Illicit Drug Use OR | Criminal Behavior OR |
|---|---|---|---|---|---|---|---|
| Model 1 | |||||||
| Gendera | .00 | −.72*** | −.27*** | −.11*** | −.01 | .78* | 3.89*** |
| Number of Missing CR Variables | .05*** | .00 | .04** | .05*** | .05*** | 1.10** | 1.04 |
| CR Linear | .12*** | .07* | .12*** | .13*** | .18*** | 1.65** | 1.29* |
| CR Quadratic | .01 | −.03 | .00 | .04 | −.04 | .92 | .97 |
| Model 2 | |||||||
| Gendera | .02 | −.72*** | −.29*** | −.11*** | −.02 | .72* | 3.62*** |
| Number of Missing CR Variables | .05*** | .00 | .04** | .05*** | .05*** | 1.10** | 1.04 |
| CR Linear | .15** | .07 | .08 | .15** | .14** | 1.43 | 1.08 |
| CR Quadratic | −.01 | −.03 | .03 | .01 | −.01 | .97 | 1.02 |
| CR Linear X Gender | −.04 | .00 | .07 | −.03 | .06 | 1.38 | 1.29 |
| CR Quadratic X Gender | .03 | .00 | −.05 | .04 | −.06 | .89 | .92 |
Note. N = 6963. CR = Cumulative Risk. β = standardized regression coefficients. OR = Odds Ratio. Girls = 0 and Boys = 1.
p < .05.
p < .01.
p < .001.
The CR Index that ranged from 0 to 3 was used in these analyses.
Because Gender is a binary independent variable, the STDY Standardization values are reported for all continuous outcome variables (Muthén & Muthén, 1998 – 2012).
Discussion
A rich tradition of research has provided support for positive relations between cumulative risk and a variety of adverse psychosocial outcomes, but relatively few studies have directly tested the additive assumption underlying these analyses (Evans et al., 2013). This study examined whether a linear or nonlinear relation best captured the association between prenatal/birth cumulative context risk and the psychosocial adjustment of adolescents. Sixteen year longitudinal data from the NFBC1986 were used. Several features of the NFBC1986 made it well-suited to answer questions about the relation between cumulative risk and youth adjustment. Specifically, few studies of cumulative risk use large, general population samples and most studies have used samples from the United States. The measures of contextual risk used in the study were obtained around the time of the child’s birth. This provided a unique opportunity to isolate the influence of early contextual risk on later adolescent psychosocial adjustment without confounding the relation with either contextual risk that occurred later in development or early adolescent adjustment. Moreover, 16 years separated the assessment of cumulative risk from the measurement of adolescent psychosocial problems, which allowed the investigation of long-term effects.
Results revealed that the majority of participants did not experience any contextual risk factors around the time of the child’s birth (58.1%). Higher levels of risk have been reported in other large population-based studies. For example, Hall et al. (2010) in a “broadly representative sample” from the United Kingdom reported the mean of their cumulative risk index to be 1.82 (11 risk factors were included in the index). In addition, the distribution of risk factors in a cohort of second graders in a large urban public school in the United States was as follows: 0 = 16.5%, 1 = 28.1%, 2 = 26.3%, and 3 or more = 26.1% (Rouse & Fantuzzo, 2009). Relatedly, findings indicated that only a small percentage of individuals in NFBC1986 sample (15.2%) were exposed to multiple risk factors. The relatively low levels of prenatal/birth cumulative risk and comparatively small co-occurrence rate among risk factors in our sample was surprising. One possible explanation for the comparatively low levels of cumulative risk observed in the NFBC1986 data is the health-related resources available to children and families in Finland. To test this possibility, research is needed that directly investigates whether making health-related resources such as prenatal care more available to families does indeed reduce children’s exposure to risk factors.
Findings indicated that even in this low-risk sample cumulative contextual risk assessed around the time of children’s birth was related to seven different psychosocial outcomes 16 years later. The form of this relation depended on how the cumulative risk index was calculated. Evidence of a nonlinear, quadratic relation was found for two outcomes (Heavy Drinking and Illicit Drug Use) when using a cumulative risk index that ranged from 0 to 5. The quadratic relations observed were consistent with a saturation effect (i.e., differences at lower levels of risk that plateaued at higher levels; Horan & Widom, 2015). One possible explanation for the saturation effects is related to measurement. Specifically, a potential problem with the cumulative risk index that ranged from 0 to 5 in our low-risk sample is that small percentages of participants experienced the highest levels of risk (less than 2% experienced 4 or more) and those individuals could exert an undue influence on the pattern of associations. We found this to be the case. When we truncated the risk variable at three or more, the highest risk group had about 5% of the sample. Findings from analyses using the cumulative risk variable truncated at three provided support for a linear relation between the cumulative risk index and all seven psychosocial outcomes. In our low-risk sample, the truncated at three approach for calculating the cumulative risk index seems most defensible given the small percentages of participants at the highest levels for the other versions of the cumulative risk variable.
The mixed results concerning linear and nonlinear effects evident in the cumulative risk literature could be explained by differences across studies in the distribution of risk factors at the highest levels of risk and, relatedly, whether low- or high-risk samples were used. Our results suggest that a clearer understanding of the relations between cumulative risk and psychosocial outcomes can be obtained if findings from future research are interpreted with the distribution of participants across the range of risk factors experienced in mind as well as the general risk level of samples.
As noted, findings using the cumulative risk index truncated at three indicated there was a linear, additive relation between cumulative risk and each psychosocial outcome. It thus appears that in this low-risk sample each additional risk factor is related to higher levels of difficulties and the pattern was similar for all of the psychosocial outcomes. Each additional risk factor seems to diminish children’s ability to cope with environmental demands. It is important to note that the magnitude of the relation between cumulative risk and psychosocial outcomes during adolescence was small. Future research is needed to identify factors at different developmental periods that mediate or moderate the association between early contextual risk and youth outcomes. This may be particularly important for internalizing symptoms given that the smallest association between cumulative risk and psychosocial outcomes was observed for internalizing problems.
Findings indicated that gender did not moderate the relation between cumulative risk and any of the adolescent psychosocial outcomes. These findings are inconsistent with prior research that suggests the relation between cumulative risk and youth outcomes is different for boys and girls (Evans et al., 2013) and that suggests the gender differences can be complex (e.g., a linear relation between cumulative risk and delinquency for boys and a quadratic relation for girls; Wong et al., 2013). The reasons why gender moderation was not found in this study are unclear. One possibility is that cumulative risk was assessed at birth in the current research and during other developmental periods in studies in which gender moderation was found (e.g., early adolescence in Buehler & Gerard, 2013; retrospective reports of family history prior to 18 in Horan & Widom, 2015; multiple developmental periods in Wong et al., 2013). Additional research is needed that directly examines whether the cumulative risk-youth outcomes associations depend on the developmental period in which cumulative risk is assessed. It is also important to note that Mason et al. (2016) using the same NFBC1986 data found that cumulative risk at the prenatal/birth period was a significant positive predictor of adolescent risky sexual behavior for girls but not boys. Thus, in the NFBC1986 data gender was only found to be a moderator of the association between cumulative risk and risky sexual behavior and not other psychosocial outcomes (Mason et al., 2016).
Results from this study should be interpreted in the context of several limitations. First, the NFBC1986 dataset did not include information on some established early contextual risk factors including neighborhood disadvantage and parental psychological functioning (other than substance use). Second, all but one of the adolescent outcomes measures (Criminal Behavior was the exception) were obtained from adolescent self-reports. Importantly, the assessment timeframes for many of the adolescent psychosocial outcomes were different from each other (e.g., current cigarette use and lifetime illicit drug use). This suggests that the findings are stable given that a consistent pattern of results was observed across outcomes. Third, we had missing data on indicators of contextual risk. To attempt to address this issue, we included a variable that represented the number of missing contextual risk indicators as a covariate in all analyses. Future research that uses the cumulative risk framework should more directly investigate the implications of missing data for interpreting results. Relatedly, we had some attrition from the prenatal/birth period to adolescence. Attrition analyses indicated that the sample for the current study was at slightly lower risk than the birth cohort. Finally, the measure of heavy drinking in the NFBC1986 study was operationalized as 6 or more drinks for boys and 4 or more drinks for girls on the same occasion. This operational definition is different than the well-established 5 or more drinks for men and 4 or more drinks for women criteria (Wechsler & Nelson, 2001). The slightly higher threshold for boys (6 drinks compared to 5 drinks) in this study could have influenced the relation between cumulative risk and heavy drinking.
Our findings highlight the potential value of using the cumulative contextual risk framework for identifying children at birth who are at risk for a range of poor psychosocial outcomes during adolescence. The assessment of contextual risk factors is relatively low-cost in terms of financial expenses, time, personnel, and training. Given that the current results support an additive relation between cumulative risk and later psychosocial outcomes, prevention efforts that target families who are experiencing both lower- and higher-levels of prenatal contextual risk could translate to reduced levels of adjustment difficulties among adolescents. The value of the cumulative risk framework is that it likely captures children who are at elevated risk better than a focus on one specific risk factor in isolation.
Acknowledgments
The analyses were supported by National Institute on Drug Abuse (NIDA), National Institutes of Health, Grant # R01 DA038450. NFBC1986 has received support from the Academy of Finland (#268336), the European Commission (EURO-BLCS, Framework 5 award QLG1-CT-2000-01643), and the US National Institutes of Health (NIMH) (5R01MH63706:02). The preparation of this manuscript was also supported by the Institute of Education Sciences (IES), U.S. Department of Education, through Grant R324B110001. The content is solely the responsibility of the authors and does not necessarily represent the official views of the funding agencies. The funders played no role in the research design, data collection, analysis, or writing and submission process. The authors would like to thank Dr. Anja Taanila for her valuable contributions to the project.
Footnotes
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
References
- Achenbach TM. Manual for the Child Behavior Checklist/4-18 and 1991 Profile. Burlington, VT: University of Vermont Department of Psychiatry; 1991. [Google Scholar]
- Appleyard K, Egeland B, van Dulmen MHM, Sroufe LA. When more is not better: The role of cumulative risk in child behavior outcomes. Journal of Child Psychology and Psychiatry. 2005;46:235–245. doi: 10.1111/j.1469-7610.2004.00351.x. [DOI] [PubMed] [Google Scholar]
- Buehler C, Gerard JM. Cumulative family risk predicts increases in adjustment difficulties across early adolescence. Journal of Youth and Adolescence. 2013;42:905–920. doi: 10.1007/s10964-012-9806-3. [DOI] [PubMed] [Google Scholar]
- Cicchetti D. Developmental psychopathology, volume 1, theory and method. 3. Hoboken, NJ: Wiley; 2016. [Google Scholar]
- Copeland WE, Adair CE, Smetanin P, Stiff D, Briante C, Colman I, Fergusson D, Horwood J, Poulton R, Costello EJ, Angold A. Diagnostic transitions from childhood to adolescence to early adulthood. Journal of Child Psychology and Psychiatry. 2013;54:791–799. doi: 10.1111/jcpp.12062. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Day J, Ji P, DuBois DL, Silverthorn N, Flay B. Cumulative social-environmental adversity exposure as predictor of psychological distress and risk behavior in urban youth. Child & Adolescent Social Work Journal. 2016;33:219–235. doi: 10.1007/s10560-015-0421-5. [DOI] [Google Scholar]
- Evans GW, Li D, Sepanski Whipple S. Cumulative risk and child development. Psychological Bulletin. 2013;139:1342–1396. doi: 10.1037/a0031808. [DOI] [PubMed] [Google Scholar]
- Felitti VJ, Anda RF, Nordenberg D, Williamson DF, Spitz AM, Edwards V, … Marks JS. Relationship of childhood abuse and household dysfunction to many of the leading causes of death in adults. The Adverse Childhood Experiences Study. American Journal of Preventive Medicine. 1998;14:245–258. doi: 10.1016/S0749-3797(98)00017-8. [DOI] [PubMed] [Google Scholar]
- Flouri E. Contextual risk and child psychopathology. Child Abuse & Neglect. 2008;32:913–917. doi: 10.1016/j.chiabu.2007.12.008. [DOI] [PubMed] [Google Scholar]
- Hall JE, Sammons P, Sylva K, Melhuish E, Taggart B, Siraj-Blatchford I, Smees R. Measuring the combined risk to young children’s cognitive development: An alternative to cumulative indices. British Journal of Developmental Psychology. 2010;28:219–238. doi: 10.1348/026151008X399925. [DOI] [PubMed] [Google Scholar]
- Hallgren KA, Witkiewitz K. Missing data in alcohol clinical trials: A comparison of methods. Alcoholism: Clinical and Experimental Research. 2013;37:2152–2160. doi: 10.1111/acer.12205. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Horan JM, Widom CS. Cumulative childhood risk and adult functioning in abused and neglected children grown up. Development and Psychopathology. 2015;27:927–941. doi: 10.1017/S095457941400090X. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Hurtig T, Ebeling H, Taanila A, Miettunen J, Smalley SL, McGough JJ, Loo SK, Järvelin MR, Moilanen IK. ADHD symptoms and subtypes: Relationship between childhood and adolescent symptoms. Journal of the American Academy of Child and Adolescent Psychiatry. 2007;46:1605–1613. doi: 10.1097/chi.0b013e318157517a. [DOI] [PubMed] [Google Scholar]
- January SAA, Mason WA, Savolainen J, Solomon S, Chmelka MB, Miettunen J, Veijola J, Moilanen I, Taanila A, Järvelin MR. Longitudinal pathways from cumulative contextual risk at birth to school functioning in adolescence: Analysis of mediation effects and gender moderation. Journal of Youth and Adolescence. 2017;46:180–196. doi: 10.1007/s10964-016-0560-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Järvelin MR, Hartikainen-Sorri AL, Rantakallio P. Labour induction policy in hospitals of different levels of specialisation. British Journal of Obstetrics & Gynaecology. 1993;100:310–315. doi: 10.1111/j.1471-0528.1993.tb12971.x. [DOI] [PubMed] [Google Scholar]
- Mason WA, January SA, Chmelka MB, Parra GR, Savolainen J, Miettunen J, Marjo-Riitta Järvelin M, Taanila A, Moilanen I. Cumulative contextual risk at birth in relation to adolescent substance misuse, conduct problems, and risky sex: General and specific predictive associations in a Finnish Birth Cohort. Addictive Behaviors. 2016;58:161–166. doi: 10.1016/j.addbeh.2016.02.031. [DOI] [PMC free article] [PubMed] [Google Scholar]
- McLaughlin KA, Green JG, Gruber MJ, Sampson NA, Zaslavsky A, Kessler RC. Childhood adversities and first onset of psychiatric disorders in a national sample of adolescents. Archives of General Psychiatry. 2012;69:1151–1160. doi: 10.1001/archgenpsychiatry.2011.2277. [DOI] [PMC free article] [PubMed] [Google Scholar]
- McLaughlin KA, Sheridan MA. Beyond cumulative risk: A dimensional approach to childhood adversity. Current Directions in Psychological Science. 2016;25:239–245. doi: 10.1177/0963721416655883. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Merikangas KR, He J, Burstein M, Swanson SA, Avenevoli S, Cui L, Benjet C, Georgiades K, Swendsen J. Lifetime prevalence of mental disorders in U.S. adolescents: Results from the National Comorbidity Survey Replication-Adolescent Supplement. Journal of the American Academy of Child & Adolescent Psychiatry. 2010;49:980–988. doi: 10.1016/j.jaac.2010.05.017. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Miettunen J, Murray GK, Jones PB, Maki P, Ebeling H, Taanila A, Joukamma M, Savolainen J, Tormanen S, Jarvelin MR, Veijola J, Moilanen I. Longitudinal associations between childhood and adulthood externalizing and internalizing psychopathology and adolescent substance use. Psychological Medicine. 2014;44:1727–1738. doi: 10.1017/S0033291713002328. [DOI] [PubMed] [Google Scholar]
- Muthén LK, Muthén BO. Mplus User’s Guide. 7. Los Angeles, CA: Muthén & Muthén; 1998–2012. [Google Scholar]
- Muthén LK, Muthén BO, Asparouhav T. Estimator choices with categorical outcomes. 2015 Retreived from https://www.statmodel.com/faq.shtml.
- Rouse HL, Fantuzzo JW. Multiple risks and educational well-being: A population–based investigation of threats to early school success. Early Childhood Research Quarterly. 2009;24(1):1–14. doi: 10.1016/j.ecresq.2008.12.001. [DOI] [Google Scholar]
- Rutter M. Protective factors in children’s responses to stress and disadvantage. In: Kent MW, Rolf JE, editors. Primary prevention of psychopathology. Vol. 3. Hanover, NH: University Press of New England; 1979. pp. 39–74. [PubMed] [Google Scholar]
- Shonkoff JP, Garner AS Committee on Psychosocial Aspects of Child and Family Health, Committee on Early Childhood, Adoption, and Dependent Care, & Section on Developmental and Behavioral Pediatrics. The lifelong effects of early childhood adversity and toxic stress. Pediatrics. 2012;129:e232–46. doi: 10.1542/peds.2011-2663. [DOI] [PubMed] [Google Scholar]
- Wechsler H, Nelson TF. Binge drinking and the American college study: What’s five drinks? Psychology of Addictive Behaviors. 2001;15:287–291. doi: 10.1037//0893-164X.15.4.287. [DOI] [PubMed] [Google Scholar]
- Wong TML, Loeber R, Slotboom A, Bijleveld CCJH, Hipwell AE, Stepp SD, Koot JM. Sex and age differences in the risk threshold for delinquency. Journal of Abnormal Child Psychology. 2013;41:641–652. doi: 10.1007/s10802-012-9695-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Zegers-Hochschild F, Adamson GD, de Mouzon J, Ishihara O, Mansour R, Nygren K, Sullivan E, van der Poel S International Committee for Monitoring Assisted Reproductive Technology & World Health Organization. Human Reproduction. 2009;24:2683–2687. doi: 10.1093/humrep/dep343. [DOI] [PubMed] [Google Scholar]