A Longitudinal Analysis of Selected Risk Factors for Childhood Psychopathology

A Longitudinal Analysis of Selected Risk Factors for Childhood Psychopathology

A Longitudinal Analysis of Selected Risk Factors for Childhood Psychopathology CARMEN NOEMI VELEZ, PH.D., JIM JOHNSON, PH.D., AND PATRICIA COHEN, PH...

3MB Sizes 0 Downloads 48 Views

A Longitudinal Analysis of Selected Risk Factors for Childhood Psychopathology CARMEN NOEMI VELEZ, PH.D., JIM JOHNSON, PH.D.,

AND

PATRICIA COHEN, PH.D.

Abstract. Eight-year and two-year longitudinal analyses ofprevalence and riskfactors foradultand adolescent psychopathology were carried out for a random sample of 776 children drawn from households in New York State. A 17.7% prevalence rate for one or more DSM-lIl disorders was estimated. In general, the risk factors studied were more related to externalizing than internalizing diagnoses. Risk factors assessed over 8-year and 2year intervals were very similar. J. Am. Acad. Child Adolesc. Psychiatry, 1989, 28, 6:861-864. Key Words: epidemiology, childhood psychopathology, longitudinal, risk factors, prevalence. The New York Child Longitudinal Study is an ongoing longitudinal study of children and their families begun in 1975. In 1985 a third wave of data was collected and more than 800 families were still enrolled. The authors hope that this brief report will lead the reader to pursue extended discussions of specific results reported elsewhere.

Data Collection The first wave of data, collected in 1975, consisted of information obtained from structured interviews with the mothers. It included questions concerning the behavior and problems of the children as well as questions about family and living conditions, childrearing practices, mother's pregnancy, and the health history of the child. Follow-up interviews included the Diagnostic Interview Schedule for Children (Costello et aI., 1984) which was given to both the child (DISC) and mother (DISC-P). Both the parent and the child interviews were conducted simultaneously by lay interviewers in the respondents' homes. The interviewer selection, training and data collection procedures have been described in detail elsewhere (Lewis et al., 1985).

Method Subjects The data for these analyses come from a sample of child caretakers (preponderantly mothers) first interviewed in 1975 in two upstate counties of New York State. The sample was selected at random from census tracts in these two counties. Nine hundred seventy-six mothers, or 86% of the eligible families with at least one child between ages I and 10, were originally interviewed regarding one randomly-sampled child within their households. Two subsequent follow-ups have been done on this original sample: in 1983 and again in 1985. In addition to the mothers, children were interviewed at follow-up. We have succeeded in reinterviewing about 80% of the original sample. The demographic characteristics of the sample of families interviewed in the follow-up are presented in Table I. The children were ages 9 to 18 years, their mean age was 13.7 years and there was an even sex distribution. Except for an overrepresentation of Catholic families and an under-representation of black and Hispanic families, both accurately reflecting the sampled area; the sample composition is broadly representative of families with children living in the eastern United States.

Measures A complete description of the algorithms used to score the diagnostic instruments is given elsewhere (Cohen et al., 1987). In brief, a two-stage procedure was followed in scoring diagnoses. At Stage I, initial criteria for diagnosis were considered to be met if either the mother or the child responded positively to items matching the minimum number of DSM-llI-R criteria, including duration criteria. In the second stage of the procedure all items considered to be relevant to each diagnosis were combined into additive scales. Sample based norms (mean and standard deviation) were obtained from these scales and used to establish severity criteria. In this second stage all children satisfying Stage I diagnostic criteria were considered to be at least "possible" cases. All children satisfying Stage I and scoring at least one standard deviation above the symptom scale mean were considered to be "probable" cases. All children satisfying Stage I and scoring two standard deviations above the mean in the continuous measure were considered to be "severe" cases. In these analyses we have used the "probable" diagnostic levels as the dependent variable.

Accepted November 16. 1988. From the New York State Psychiatric Institute, 722 West 168th Street. New York. New York. Dr. Velez is Associate Professor at the School of Public Health. University of Puerto Rico. Drs. Velez. Johnson. and Cohen are Research Scientists at the New York Psvchiatric Institute. Dr. Velez was Co-Project Director and Dr. Cohen' is Principal Investigator of the Children in the Community Study ofwhich this is a part. This research was supported by NIMH. Grant MH36971 (Epidemiology of Persistent Child Psychopathology). Reprint requests and correspondenceto Dr. Cohen. New York State Psychiatric Institute. 722 West 168th Street. New York. NY 10032.

Data Analyses In terms of the risk factors considered here, simple rules were followed for selecting cutoffs to estimate the relative risks. For risk factors measured on a continuous scale such as income, the relative risk was calculated by comparing children

0890-8567/89/2806-0861 $02.00/0©1989 by the American Academy of Childand Adolescent Psychiatry. 861

862

VELEZ ET AL. TABLE

I. Characteristics of the 1983 Sample of Children Ages 9 to 18 (N = 776)

Mean age Males Whites Catholics Divorced by 1975 Mothers education (mean years) Fathers education (mean years) Mean no. of children in family Mean family income

13.7 50% 92 %

55% 8.7% 12.6 12.9 2.4 23,800

one standard deviation below the mean with children one standard deviation above the mean on the risk factor. For categorical variables the relative risks for selected contrasted groups are presented. Logistic regression analysis was used to determine risk, adjusting for age, sex, and socioeconomic status (SES) where appropriate. Two different longitudinal analyses will be presented . The first one represents an assessment of the long-term risk factors. There were 8 years between the assessment of the risk factors and the diagnoses. Thus, children were between I and 10 years old when the risk factors were assessed in 1975 and ages 9 to 18 when the diagnostic information was collected in 1983. The second set of longitudinal analyses examined a shorter interval (two years) between the assessment of the risk factors and the diagnoses. Children were aged 9 to 18 at risk assessment (1983) and II to 20 at diagnosis (1985).

TABLE

2. Time 2 and Time 3 PrevalencesofDSM-I/l-R Diagnoses by Age Time 2

Ages Diagnosis Attention deficit (ADD) Conduct disorder(CDD) Oppositional disorder (OPP) Overanxiety (OVA) Separation anxiety (SEP) Major depression (MOD) One or more severe diagnoses

9 to 12

13 to 18

I I to 14

15 to 20

16.6° 53 11.9" 38

9.9 45 11.6 53

12.8 41 10.9" 35

6.8 31 9.9 45

15.6 50 19.1 61

18.6 85 12.7" 58

22.5 72 9.7 31

14.3 65 8.6" 39

(N)

25.6 82 2.5 8

6.8 31 3.7" 17

15.3 49 2.5" 8

4.4 20 3.1 14

(%) (N) N

19.4 62 320

(%) (N)

(%) (N)

(%) (N)

(%) (N)

(%) (N)

(%)

Prevalence ofSelected DSM-l/I-R Diagnoses

The prevalence at each follow-up for the six main diagnoses selected for analyses are presented in Table 2. Table values are provided for estimates of prevalence of one or more diagnoses, made at the "severe" level, and including substance abuse disorders. Conduct disorders (COD), attention deficit disorders (ADD), oppositional disorders (OPP), over-anxious disorders (OYA), separation anxiety (SEP), and major depression disorders (MOD) were selected for analyses because they were either fairly prevalent conditions in the sample studied or serious enough to warrant further analysis. In general. age and sex specific rates are well within those expected from previous epidemiological and clinical studies. ADD and OYA were found to decline with age. COD and OPP were found to reach a peak in middle adolescence and then show a decline. OYA showed a sex-specific decline with age for boys so that the expected female predominance was confined to the older groups in both data sets. MOD was rare, increasing slightly with age and showing a greater prevalence in girls. The externalizing disorders. except for OPP , were more common in boys. No sex difference was found for SEP. Longitudinal Analysis: I. Eight- Year Interval

Table 3 presents the first and longer interval longitudinal analysis of selected risk factors for childhood psychopathol-

16.4 75 456

15.6 50 320

16.0 73 456

" Significantly more males than females. " Significantly more females than males.

TABLE

3. Relative Risk of Factors Measured at Ages 1 to 10 on DSM-I/l-R Diagnoses Measured at Ages 9 to 1S" Time 2 Diagnoses Externalizing

Time I Factors Results

Time 3

Demographics Low SES" Low income Low mother education Low father education Family Structure Never married/ SES Divorced at Time I Ethnicity Nonwhite/SES Parental characteristics Sociopathy/SES Pregnancy problems

Internalizing

COD ADD OPP OVA SEP MOD 2.35 2.74

2.67 2.60

2.03 2.20

1.55 4.45 2.23

1.98

2.57

2.05

3.71

1.70

4.97

2.52

3.65

3.74

2.43 2.87 1.79

1.49

1.66 1.63

Total sample controlling for age and sex, and SES when indicated, estimated by adjusted odds . " SES. socioeconomic status. a

ogy. Only statistically significant odds ratios are shown in the tables. The first analyses examined some socio-structural risk factors beginning with low SES. The analysis indicates that low SES represented a statistically significant risk factor for all externalizing diagnoses as well as for SEP. Low SES children

863

EPIDEMIOLOGY: LONGITUDINAL ANALYSIS

developed SEP disorders at twice the rate of high SES children. This relationship has also been reported in clinically based studies. When the components of the SES index were evaluated individually it is interesting to note that maternal education represented a rather general risk factor associated with five of the six diagnoses studied, but was not related to overanxiety. Income was found to be related to all the externalizing diagnoses and to SEP as well. Father's education on the other hand, was related only to ADD and to SEP. In order to assess the relative and independent importance of the components of SES, income, mother's education and father's education were entered simultaneously into the regression equation. In these analyses income retains its predictive power for all externalizing disorders. For SEP however, parental education remained a significant predictor, while income was no longer significant. Family structure was the next potential risk evaluated. When the mother reported never having been married to the father of the child, children were almost 2.5 times more likely to develop CDD. This variable represented a risk for OPP. Families in which divorce has occurred before the child was 10 years old represented only a marginal risk factor net of SES. Ethnicity (being black) was found to represent a risk for SEP even after controls for SES were included . Some parental characteristics were also analyzed as poten-

tial risk factors for childhood psychopathology. Parental sociopathy measured as a parental drug , alcohol , or police problem early in the child's life was found to be a risk factor for CDD and OPP, even after SES was controlled for in the analysis. A summary count of pregnancy problems was the next variable considered. It was found to be related to OVA. Elsewhere we have analyzed this variable in more detail and found it to be an important risk factor for all six types of psychopathology when measured continuously (Cohen et al., 1989c). In these further analyses, risks associated with pregnancy problems were found to be robust even when other crucial factors were considered (e.g., intelligence and family structure). In summary, the risk factors as measured in early and middle childhood were found to be related more to the externalizing than to the internalizing disorders in late childhood and adolescence . Among the components of SES, low maternal education was found to be a general long-term risk factor for childhood psychopathology . Low paternal education, on the other hand. was found to be related only to ADD and SEP. Among the internalizing diagnoses, SEP was found to be related to more of the risk factors examined than were the other internalizing diagnoses. Overanxiety proved to be unrelated to most of the factors studied, showing a relationship only with pregnancy problems .

TABLE 4. Relative Risk of Factors Mea sured at Axes 9 10 18 on DSM-Ill-R Diagnoses M easured at Ages II 1020" Time 3 Diagnoses Externalizing Time 2 Factors Socio-Structural Demographics Age/sex Sex/age Low SES Low income Low mother education Low father education Income/mo. ed/fa. ed Residential stabilit y Frequent moves Recent moves Famil y structure Mother only/SES Stepfather/SES Race Nonwhite /SES Parental characteristics Sociopath y Father emotional problem s/SES Mother emotional problems/SES Both parents with emot ional problems Child characteristics Ever repeated grade Mental health treatment by Ti me 2 Stressful life events (7 or more) since Time 2 vs. 2 or less Q

h

Internal izing

CDD

ADD

OPP

0.98 1.39 2.26 2.90

0.89

0.93

1.78 h Inc.

3.79 3.15 2.78 Inc.

1.99 2.24 1.92 2.53 Inc.

OVA

SEP

MDD

0.77 0.70 1.91

0.50 5.41 3.59 4.12 3.90 Inc, Mo. Ed

1.85 2.03

1.74 2.02

2.69 1.54

1.47

1.52 2.33 2.32 5.35

3.56 6.61

2.40 2.29

3.84 2.75

1.72 2.46

1.93 2.29

2.62

3.65

2.39

2.14

1.67

1.91

Always controlling for age and sex. and SES when indicated. estimated by adjusted odds. Significant in simultaneous estimation .

864

VELEZ ET AL.

Longitudinal Analysis: II. Short-term Interval The second part of the analysis considered a shorter interval of time. In this second section the risk factors were measured when the children were 9 to 18 years old and the diagnoses made when the children were 11 to 20 years old. In this shorter interval analysis the findings concerning SES were very similar to those reported previously. That is, SES was related to all externalizing disorders, whereas it was a strong risk factor only for SEP among the internalizing disorders. When all three components of SES were entered simultaneously into the equation, income was the only factor to retain its predictive power. Maternal education, however, did show an independent effect for SEP. In terms of family structure, children in households headed by mothers were found to be at some additional risk for COD and for OPP. On the other hand, the presence of a stepfather in the home was found to be a significant risk factor for ADD and marginally for MOD. A more detailed analysis of the family structure variables found larger sex specific effects; for example, the presence of a stepfather was related to COD in boys but not in girls (Cohen et aI., 1987). As children grew older, nonwhite status lost its significance with respect to SEP and gained some significance with respect to COD. Multiple residential moves was not a risk factor for the diagnoses studied, but moved within the last 2 years was a risk factor for COD. More complete analyses and discussion of the risk associated with residential instability can be found in Cohen et al. (l989a, b). In terms of parental characteristics, sociopathy in the family (controlling for SES) was a marginal risk factor for two of the externalizing diagnoses: COD and ADD. Mothers' reports of their own mental health problems, controlling for reported paternal mental health problems and for SES, was related to OPP and MOD, whereas paternal history of mental health problems was a risk factor for OVA. Of the three child characteristics examined (ever repeating a grade, ever having mental health treatment, and stressful life events) children who had repeated a grade at some point in their school lives were more likely to have each diagnosis except SEP. As one might expect, because of its effect on school performance, the greatest risk was for ADD. Life events were assessed as risk factors retrospectively: when the children were II to 20 years old they were asked about the occurrence of23 different stressful life events during the previous 2 years. Based on the child's report, a continuous measure of number oflife events was constructed and a cutoff of seven events was selected for these analyses. It was found that children who reported seven or more events in the past 2 years were almost three times as likely as children reporting fewer events to have a diagnoses of COOs and almost twice as likely to have other externalizing diagnoses (ADD and OPP). Life events were also found to represent a somewhat weaker risk factor for OVA. Further analysis of individual events indicated that problems in schools or at a job, problems with the law, and serious fights with the family were among

the most important life events associated with the externalizing diagnoses. It should be noted that these are precisely the events which are most difficult to disentangle from the diagnoses themselves. Summary and Conclusions In both the 8-year and 2-year longitudinal analyses the factors studied were more related to the externalizing than to the internalizing diagnoses, in general. This second analysis also replicated the finding of the relative independence of overanxiety disorders from the structural factors found in our earlier work (Cohen and Brook, 1987). Only pregnancy problems and parental psychiatric history were related to overanxiety, findings that could be suggestive of a more biological determination of this disorder. Among the internalizing diagnoses, separation anxiety was most frequently related to the factors considered in both the short- and long-term longitudinal analysis. Major depressive disorder was very rare in the sample studied so that there was reduced power to evaluate effects of the various risk factors. Thus, only very strong effects were identified in these analyses. Relationships observed are generally larger when continuous measures of the different disorders are used rather than the diagnostic categories employed here. Analyses are now underway to determine the mechanisms by which the structural risk factors operate by means of family interactive and relationship variables, school and neighborhood effects, and subcultural factors. Finally it was noted that the estimates of risks reported here are longitudinal, and that cross-sectional estimates based on these same data produce larger estimates. This fact needs to be taken into consideration when comparisons are made with results from other studies reported in this volume. References Cohen, P. & Brook, J. S. (1987), Family factors related to the persistence of psychopathology in childhood and adolescence. Psychiatry, 50:332-345. - - Johnson, J., Lewis, S. A., & Brook, J. S. (I 989a), Single parenthood and working mothers. In: Stress between work and family. ed. J. Eckenrode & S. Gore New York: Plenum. - - Johnson, J., Struening, E. L. & Brooke, J. S. (l989b), Family mobility as a risk factor for psychopathology in childhood. In: Epidemiology and the prevention ofmental disorders. ed. B. Cooper & T. Helgeson London: Rutledge. - - Velez, C. N., Smith, J. & Brook, J. S. (l989c), Mechanisms of the relationship of perinatal problems, early childhood illness, and psychopathology in late childhood and adolescence. Child Dev.60:70 1-709. - - Velez, C. N., Kohn, M., Schwab-Stone, M. & Johnson, J. (1987), Child psychiatric diagnosis by computer algorithms. J. Am. Acad. Child Adol. Psychiatry, 26:631-638. Costello, A. J., Edelbrock, C. S., Dulcan, M. K. & Robert Kalas. (1984), Testing of the NIMH Diagnostic Interview Schedule for Children (DISC) in a clinical population. (Contract No. RFP-DB~ 1-0027. Rockville, Md: Center for Epidemiological Studies, National Institute of Mental Health. Lewis, S. A., Gorksky, A., Hartmark, C. & Cohen, P. (1985), The reactions of youth to diagnostic interviews. J. Am. Acad. Child Adol. Psychiatry, 24:750-755.