American Journal of Epidemiology Copyright 998 by The Johns Hopkins University School of Hygiene and Public Health All rights reserved Vol. 47, No. 2 Printed in U.S.A. Risk Associated with Various Definitions of Family History of Coronary Heart Disease The Newcastle Family History Study II Jonathan S. Silberberg, ' 2 John Wlodarczyk, Jayne Fryer, Randall Robertson, and Michael J. Hensley The authors carried out a population-based case-control study to estimate the risk of an acute coronary disease event associated with various definitions of a family history of coronary heart disease (CHD). A detailed family history questionnaire was completed by 403 cases and 236 controls in Newcastle, New South Wales, Australia from 992 to 994. Odds ratios of an acute coronary disease event adjusted for proband age and sex ranged from 2.7 (95% confidence interval (Cl).8-4.) for the simplest definition (one or more first-degree relatives with CHD at any age) to 5.4 (.7-6.8) for the most stringent definition (two or more first-degree relatives with CHD before age 55 years). In a series of nested models, the authors examined the improvement in model fit as each component of the detailed family history was added. Additional information was provided by accounting for "don't know" responses, the number of affected relatives, the age of the affected relative, and whether the first-degree relative was a sibling rather than a parent. The results were similar when the data were analyzed as a cohort design with proband disease status as the exposure variable. The authors suggest that, to facilitate preventive efforts in a population, more detailed family history definitions should be used to better target high risk subjects. Am J Epidemiol 998;47:33-9. age of onset; case-control studies; coronary disease; family characteristics; genetics A strong family history of coronary heart disease (CHD) is the major predisposing factor to the development of CHD. Despite this, family history of CHD is given no more status than that of other risk factors in current guidelines for the treatment of dyslipidemia in the United States (), Europe (2), Australia (3), or New Zealand (4), and some treatment algorithms based on absolute risk (5) do not consider family history at all. When family history is included, it is rarely used precisely. Usual definitions of a positive family history (such as having a first-degree relative with CHD) do not consider all of the information available. As has been shown in the Utah population (6), the risk associated with a positive family history is Received for publication October 30, 996, and accepted for publication August 27, 997. Abbreviations: CHD, coronary heart disease; Cl, confidence interval; MONICA, Monitoring Trends and Determinants in Cardiovascular Disease; OR, odds ratio. Discipline of Medicine, Faculty of Medicine and Health Sciences, The University of Newcastle, Callaghan, Newcastle, New South Wales, Australia. 2 Cardiovascular Unit, John Hunter Hospital, Newcastle, New South Wales, Australia. Reprint requests to Dr. Jonathan Silberberg, Department of Medicine, John Hunter Hospital, Locked Bag No., Hunter Region Mail Centre, NSW 230, Australia. greater when accounting for the number of at-risk relatives or the age at which CHD first became evident. To better identify those at high risk for CHD in a typical Australian population, we sought to ascertain the prevalence of various definitions of a positive family history and to estimate the additional information provided by more complex definitions. We evaluated the risk using a traditional case-control approach and also by comparing cohorts of case and control relatives. MATERIALS AND METHODS The Family History Study was a population-based study of family history of CHD conducted in Newcastle, New South Wales, Australia, from 992 to 994. Details of the sampling frame, characteristics of participants, and the accuracy of recall of family history are described in Paper I (7) accompanying. Briefly, we enrolled suspected CHD cases among males aged under 65 years and among females aged under 70 years who were admitted to one of the Hunter region district hospitals with a suspected acute coronary disease syndrome (myocardial infarction or unstable angina) who were registered by the Newcastle MONICA project (8) (Monitoring Trends and Determinants in Cardiovascular Disease). Controls were 33
34 Silberberg et al. invited at random from the Hunter region of the New South Wales electoral roll. In the main analysis, we report the odds ratio of an acute coronary disease event based on comparison of self-reported family history between cases and controls who completed the detailed family history questionnaire. Family history definitions We defined a positive family history in several ways, according to the number of affected relatives, their relationship to the proband, and their age when CHD first became evident. Simple definitions ranged from "one or more first-degree relatives with CHD at any age" to "two or more first-degree relatives with CHD before age 55 years." Within each of these definitions, we further defined which relative was affected (sibling or parent, father or mother, brother or sister). Because few children were affected, they were excluded from the main analysis. Statistical methods Case-control approach. We performed logistic regression analysis (SAS Institute, Cary, North Carolina) with disease status (case/control) as the outcome variable. The explanatory variables were family history of CHD and proband sex and age (continuous). We included the proband's smoking status, history of high blood pressure, high cholesterol, and diabetes mellitus as covariates. We first fitted a series of mutually exclusive models and derived odds ratios for each of the family history definitions. To account for "don't know" responses, we assigned an indicator (dummy) variable for "don't know" to each family history definition. Nested models. We tested the additional information afforded by more precise definitions of family history in a series of nested models (9) to study the improvement in goodness-of-fit as each component family history term was added. For instance, the importance of age at onset of CHD in the relative was examined by comparing the model which included both "one or more affected relatives at any age" and "one or more affected relatives before age 60 years" with that including the "any age" term alone. The coefficient for the "before age 60 years" term addresses the odds ratio associated with onset before age 60 years compared with any age. Cohort approach. Khoury and Beaty (0) have proposed an alternative to treating family history as an "exposure" in case-control studies by transforming the analysis to a cohort design. We evaluated the cohort of relatives for cumulative risk of CHD, with proband disease status (case or control) as the "exposure" variable. We conducted several analyses for first-degree, second-degree, and all relatives combined as well as parents, siblings, brothers, and sisters. We fitted the models of Zeger and Liang () using the SAS macro GEE version 2.03 (SAS Institute, Cary, North Carolina) in order to estimate the within-family correlation in risk with relatives' age and sex and proband disease status in the model. We compared the risk estimates with those derived from the traditional case-control analysis. Sensitivity analyses To study the effect of misclassifying "don't know" to "no," we pooled these as the reference category. Because our validation study indicated that most misclassification of CHD was to "other heart disease" rather than "no heart disease," we also conducted the analyses based on family history of all heart disease, rather than CHD alone. In other sensitivity analyses, we used incident rather than prevalent cases and fatal CHD in relatives rather than all affected relatives. To address the impact of selection bias, we repeated the comparisons of simple family history definitions using the larger data set of all short questionnaire respondents (,08 cases and 574 controls) rather than the smaller number who completed the detailed questionnaire. RESULTS A detailed family history questionnaire was completed by 432 MONICA subjects and 248 electoral roll subjects. We excluded 29 MONICA subjects because they had a final diagnosis other than "definite" or "possible" myocardial infarction and 2 electoral roll subjects who reported a history of prior angina or myocardial infarction. The demographic particulars of the 403 cases and 236 controls who completed the detailed questionnaire were similar to those who completed the short questionnaire only, as described in Paper I (7). Cases were older and included a higher proportion of males. They more often reported prior cigarette smoking, a history of hypertension, or diabetes mellitus. Controls were more likely to be employed full time and to report an education level beyond high school. Case-control approach The odds ratios associated with various definitions of family history of CHD, adjusted for proband age and sex, are given in table. The odds ratios ranged from 5.4 (95 percent confidence interval (CI).7-6.8) for the most stringent definition (two or more first-degree relatives with CHD before age 55 years; Am J Epidemiol Vol. 47, No. 2, 998
Risk Associated with Family History of Coronary Heart Disease 35 TABLE. Odds ratios (OR) and 95% confidence interval (Cl) associated with family history of coronary heart disease (CHD) in the Newcastle Family History Study, Australia, 992-994* Family history First-degree relatives (parents, siblings) with CHD at age <55 years > with CHD at age <60 years > with CHD at any age >2 with CHD at age <55 years >2 with CHD at age <60 years >2 with CHD at any age Parents -2 with CHD at age <60 years -2 with CHD at any age Siblings (brothers and sisters) > with CHD at age <60 years > with CHD at any age Prevalence (%:) Cases Controls 28 37 76 6 0 42 2 68 22 37 4 57 2 3 2 8 53 9 6 No.not "donl know" Cases 364 364 380 326 326 346 39 353 308 35 Controls 223 223 226 207 207 23 22 28 93 98 4.3 4.5 2.7 5.4 5. 3.2 5. 2.4 3.4 3.6 "Don't know" as a dummy variable 2.6-7. 2.8-7..8-4..7-6.8 2.2-2.3 2.^4.9 2.9-9.3.6-3.6 2.0-5.9 2.3-5.8 "Doni know" pooled with "no" 4. 4.2 2.4 5. 4.8 2.7 4.6.9 3. 2.9 2.5-6.7 2.7-6.6.6-3.5.6-5.9 2.0-.4.8-^.0 2.5-8.2.3-2.7.8-5.3.9-4.6 * Because "don't know" responses varied with the components of each definition, the number of infomnative subjects differs in each analysis; the prevalence shown is that with "don't knows" included. f Odds ratio based on detailed family history questionnaire and adjusted for proband age and sex. prevalence in sample = 2 percent) to 2.7 (95 percent Cl.8-4.) for the simplest definition (one or more first-degree relatives with CHD at any age; prevalence in sample = 57 percent). Odds ratios for onset at any age were lower than for early onset, except where the affected relative was a sibling (hence closer in age to the proband). Surprisingly, the estimates for an affected father were greater than for an affected mother, but the confidence limits about the difference were wide. The odds ratios for one or more affected relatives are illustrated in figure. Nested models: improvement in fit with more detailed family history As our reference model, we chose the simplest family history definition (one or more first-degree relatives with CHD at any age) and compared the improvement in fit as further components of the family history were added. These analyses are summarized in table 2. at onset. The odds ratio associated with onset of CHD in a first-degree relative before age 60 years 00 or more first degree relatives father or more brothers - ormore -sisters in tn o to <o a) W V < of onset FIGURE. Odds ratios (and 95 percent confidence intervals) of an acute coronary disease event associated with various definitions of family history of coronary heart disease (CHD), according to age (years) at onset of CHD in relatives. Odds ratios are adjusted for proband age and sex, with "don't know" as the dummy variable. Am J Epidemiol Vol. 47, No. 2, 998
36 Silberberg et al. TABLE 2. Odds ratios (OR) and 95% confidence interval (Cl) and improvement in fit with addition of detailed family history terms in the Newcastle Family History Study, Australia, 992-994 Te moared at onset of affected first-degree relatives Onset at age <60 years vs. onset at any age Onset at age <55 years vs. onset at any age Onset at age <55 years vs. onset at age <60 years No. of affected first-degree relatives >2 vs. >, at any age >2 vs., at age <60 years "Don't know" about first-degree relatives "Don't know" as dummy variable vs. "don't know" reset to "no," at any age "Don't know" as dummy variable vs. "don't know" reset to "no," at age <60 years Relationship of relative Sibling vs. any first-degree relative, at any age Sex of relative Mother vs. either parent, at any age Sister vs. any sibling, at any age Odds ratio* 3.7 3.4.2 2.5.8 2.2.9 2.6.0.4 2.3-6.0 2.0-5.7 0.5-3..6-4.0 0.7-4.7 0.9-5. 0.9-3.9.6-4.3 0.6-.6 0.7-3. X 2 32. 24.4 0.2 8. 5.0 3.2 4.4 7.3 0.0 0.8 Improvement in model fit * Odds ratio associated with the more detailed definition when compared with the other definition listed, t d.f., degrees of freedom. d.f.t 2 2 n value <0.000 <0.000 0.65 <0.000 0.08 0.08 0.04 <0.000 0.94 0.37 compared with onset at any age was 3.7 (95 percent Cl 2.3-6.0). Onset before age 55 years did not add further to onset before age 60 years. Number of affected relatives. The odds ratios comparing two or more affected first-degree relatives with one or more affected were 2.5 (95 percent Cl.6-4.0) for onset of CHD at any age and.8 (95 percent Cl 0.7-4.7) for onset of CHD before age 60 years. Relationship of relative. The odds ratio associated with the affected relative being a sibling rather than any first-degree relative (i.e., sibling or parent) was 2.6 (95 percent Cl.6-4.3). This was particularly important in younger probands (see below). Sex of relative. The coefficients for affected mother (vs. any parent) and sister (vs. any sibling) were not significant (improvement, x* = 0.0 and 0.8, respectively). Effect modification by proband age and sex. All odds ratios associated with the family history definitions were greater at younger ages (table 3). There was little modification by gender. In younger probands, an affected sibling was significantly associated with risk even in families with a parent already affected before age 60 years (odds ratio (OR) for sibling adjusted for parent = 5.8; improvement, )f 8.4, 2 degrees of freedom (df), p < 0.000). In older probands, the improvement was less (x 2 = 3.6, 2 df, p = 0.6). Favorable family history. For the definition "no TABLE 3. Odds ratios (OR) and 95% confidence intervals (Cl) associated with family history definitions, by age and sex of the proband, in the Newcastle Family History Study, Australia, 992-994 Family history definition age No. Of cases No. of controls first-degree relatives with CHD* at any age ORf 2 first-degree relatives with CHD at arry age first-degree relatives with CHD al age <60 years 2 first-degree relatives with CHD at age <60 years siblings with CHD at any age Males <55 55 Females 55 55 28 34 39 02 78 2 68 69 3.3.8-6.0.8 0.6-5. 4.2.6-0.8 2.4 0.9-6.0 4.5.9-0.8 2. 0.7-6.0 8.6 3.-23.4 2.3.2-4.6 6.8 3.0-5.3 2.8 0.8-0. 6.6 2.6-6.8 2.4.-5.0.8.5-90.7.7 05-4.4 3..5-3.8.9 0.5-6.4 6.4 2.-9. 2.8 0.9-8.5 7.3 2.3-23.7 3.3.6-6.7 * CHD, coronary heart disease. t Odds ratio with "don't know" as a dummy variable. Am J Epidemiol Vol. 47, No. 2, 998
Risk Associated with Family History of Coronary Heart Disease 37 first-degree relative with CHD and two or more not affected by age 75 years," the odds ratio was 0.3 (95 percent CI 0.-0.7). This definition was met by 9/403 cases (2 percent) and 8/236 controls (8 percent). The more common definition "no first-degree relative with CHD and two or more not affected by age 70 years" yielded an odds ratio of 0.45 (95 percent CI 0.3-0.8). "Don't know" is significantly associated with risk of CHD. The odds ratios for those who reported "don't know" compared with those who reported "no" were different from unity for most family history definitions. For instance, "don't know whether any firstdegree relative had CHD before age 60 years" yielded an odds ratio of.9 (95 percent CI 0.9-3.9). Sensitivity analyses Misclassifying "don't know" responses. When all "don't knows" were reclassified as "no," the adjusted odds ratios associated with a positive family history were slightly reduced, ranging from 2.4 (95 percent CI.6-3.5) for "one or more first-degree relatives with CHD at any age" to 5. (95 percent CI.6-5.9) for "two or more first-degree relatives with CHD before age 55 years." Family history of all heart disease. For family history of all heart disease, the frequencies and odds ratios were similar to those for CHD. For instance, the odds ratio associated with "one or more first-degree relatives with heart disease at any age," adjusted for proband age and sex, was 2.5 (95 percent CI.7-3.8). Incident cases only. Of 403 cases who completed the detailed questionnaire, 243 (60 percent) had no prior history of CHD. All odds ratios were similar when restricted to this group. Fatal affected relatives only. Using the short questionnaire,,025 probands reported an affected firstdegree relative, with 734 first-degree relatives who had died of CHD. The odds ratio associated with one or more fatally affected first-degree relatives at any age was 2.4 (compared with 2.7 for either fatal or nonfatal). When adjusted for the age of the proband and with "don't know" responses excluded, the odds ratio was 2. (p < 0.000). Impact of selection bias: analysis of short family history only A short family history which did not include the age of affected relatives was available for,08 cases and 574 controls. Disagreement between the short and detailed family history was infrequent (leading to a change of family history status in only 6 percent of subjects). Adjusted odds ratios were slightly smaller than those for the main analysis (for one or more affected first-degree relatives at any age, OR = 2.5, 95 percent CI 2.2-2.8; for one or more affected siblings, OR = 2.7, 95 percent CI 2.3-3.). Adjustment for other risk factors In multivariate analyses, the major CHD risk factors were all significantly associated with an acute coronary disease event. Odds ratios were 2.0 (95 percent CI.3-3.0) for having ever smoked, 6.8 (95 percent CI 2.4-9.5) for diabetes mellitus, 2.4 (95 percent CI.6-3.6) for high blood pressure, and 2.9 (95 percent CI.9-4.5) for high cholesterol. When these risk factors were included in the model, the odds ratios for all family history definitions were slightly reduced (e.g., for one or more first-degree relatives with CHD at any age, OR = 2.3, 95 percent CI.5-3.6, and for one or more first-degree relatives with CHD before age 60 years, OR = 4.3, 95 percent CI 2.6-7.0). Cohort approach The number of relatives, within-family correlations in risk, and odds ratios derived in each analysis using the cohort approach are shown in table 4. In firstdegree relatives (excluding children), the odds ratio was 2.3 (95 percent CI.6-3.); in siblings, 3.0 (95 percent CI.9-4.7); and in second-degree relatives,.5 (95 percent CI.0-2.2). The odds ratio for gender was.8 and that for age.07 (per year). The largest intraclass correlation in risk was 0.9 (for male siblings); correlations of 0.05 to 0.0 were more usual. DISCUSSION The estimates of coronary disease risk associated with a positive family history of CHD that we obtained are all similar to those reported by Hunt et al. (6) in the Utah population. When "don't knows" were reclassified as "no," the odds ratios remained large. This confirms that a detailed family history of CHD can be used to identify subjects at high or low risk of CHD even in a population where a high proportion of "don't know" responses is encountered. The prevalence of the various family history definitions we observed in the control series cannot be extrapolated to the general population without considering the selection and age stratification of our sample. Although the participation rate was low, we are reasonably sure that our risk estimates reliably reflect the source population. The analysis including 76 percent of eligible MONICA subjects and 62 percent of eligible electoral roll subjects based on a short family history gave similar estimates to the main analysis. We estimated the selection and recall biases explicitly. Even under a "worst case" scenario, the risk associated Am J Epidemiol Vol. 47, No. 2, 998
38 Silberberg et al. TABLE 4. Odds ratios and 95% confidence intervals (Cl), by model, under the cohort approach in the Newcastle Family History Study, Australia, 992-994* Model No. of relatives Parents and siblings,56 No. of families 57 Within-family correlation In disease 0.07 Model terms Sex Odds ratio, robust variance.06.6 2.3.05-.08.2-2..6-3. Male siblings 583 354 0.85.08 2.9.05-.0.6-5.3 Female siblings 63 338 0.096.08 3.4.05-.0.7-6.7 Second-degree relatives 854 293 0.46 Sex.04.9.5.02-.05.4-2.6.0-2.2 with a strong family history of CHD remains substantial. When using the odds ratios derived from the casecontrol analysis as measures of relative risk, one must be mindful of biases intrinsic in the method. Khoury and Flanders (2) evaluated the impact of several family parameters. They showed that even when there are no differences in family size or age profile, the odds ratio from a case-control analysis may be biased in a positive (unfavorable) direction when the disease is prevalent among relatives or when the correlation in risk between relatives is small (as applies in the present study). Other factors which may introduce bias are unequal numbers or unequal age distribution of case and control relatives and early onset of disease (these factors are either neutral or are slightly favorable in this study). The magnitude of bias inherent in the case-control analysis can be seen by comparing odds ratios from the two approaches. When adjusted for age and sex, the definition "one or more first-degree relatives with CHD" yielded odds ratios of 2.7 (case-control) and 2.3 (cohort). Other odds ratios were similarly inflated, and the conclusions were generally unaltered. The high prevalence of CHD among relatives (approximately 30 percent) contributes the most to bias in the case-control analysis. Because CHD is more common in males, the bias will be stronger for fathers and brothers. This is likely to account for the finding that an affected father conferred higher risk than an affected mother (figure ). An alternative way to account for family size, age, and risk status is to derive family history scores or indexes which compare the family experience with that expected from population CHD rates. This enables the case-control format to be retained. We have applied published scores to these data and have derived two of our own, as reported elsewhere (3). Our data show clearly the value of looking beyond first-degree "yes/no" definitions. In addition to allowing for the number of affected relatives and their age at disease onset, we found the fit to be improved when distinguishing an affected sibling from an affected parent, particularly in younger probands. This observation is consistent with the findings of a major study of familial CHD in twins (4) but has not usually been considered in CHD risk estimation. Published estimates of risk associated with a family history of CHD have varied with the definition used and also with the inclusion of covariates in the model (5). It has been usual to pursue the effect of family history independent of other risk factors such as diabetes mellitus, high blood pressure, or high cholesterol. In this way, adjusted odds ratios of between.3 and 3.0 have been reported (5). It has been observed that running in families are several well-defined metabolic disorders whose major manifestation is premature atherosclerosis. As these phenotypes are better defined and entered in the model, the "independent" effect of family history will disappear entirely. In the population, a strong family history indicates how vigorously causal phenotypes should be sought. Even in those who do not have a clearly defined metabolic anomaly, family history serves as a marker of an underlying familial susceptibility to risk factors (2) and perhaps a stronger case for treatment. We recommend that evaluation of a family history of CHD includes the number of affected relatives and whether CHD became manifest before age 60 years. In younger probands, identifying those with an affected sibling appears particularly important. Am J Epidemiol Vol. 47, No. 2, 998
Risk Associated with Family History of Coronary Heart Disease 39 ACKNOWLEDGMENTS This paper was funded by Commonwealth Department of Human Services and Health (RADGAC) grant no. HS 335. REFERENCES. Expert Panel on Detection, Evaluation and Treatment of High Blood Cholesterol in Adults. Summary of the Second Report (Adult Treatment Panel II). JAMA 993;269:305-23. 2. Pyoralla K, De Backer G, Graham I, et al. Prevention of coronary heart disease in clinical practice: recommendations of the Task Force of the European Society of Hypertension. Atherosclerosis. 994;0:2-6. 3. National Heart Foundation of Australia. Guide to plasma lipids for doctors. Woden, ACT, Australia: National Heart Foundation of Australia, 995. 4. Dyslipidaemia Advisory Group on behalf of the Scientific Committee of the National Heart Foundation of New Zealand. 996 National Heart Foundation clinical guidelines for the assessment and management of dyslipidaemia. NZ Med J 996;09:224-3. 5. Haq IU, Jackson PR, Yeo WW, et al. Sheffield risk and treatment table for cholesterol lowering for primary prevention of coronary heart disease. Lancet 995;346:467 7. 6. Hunt SC, Willams RR, Barlow GK. A comparison of positive family history definitions for defining risk of future disease. J Chronic Dis 986;39:809-2. 7. Silberberg J, Wlodarczyk J, Fryer J, et al. Correction for biases in a population-based study of family history and coronary heart disease. The Newcastle Family History Study I. Am J Epidemiol 998; 47:23-32. 8. Tunstall-Pedoe H, Kuulasmaa K, Amouyel P, et al. Myocardial infarction and coronary deaths in the World Health Organization MONICA Project. Circulation. 994;90:583-62. 9. Miettinen OS. Theoretical epidemiology. New York: John Wiley & Sons, 985:222. 0. Khoury MJ, Beaty TH. Applications of the case-control method in genetic epidemiology. Epidemiol Rev 994;6: 34-50.. Zeger S, Liang KY. Longitudinal data analysis for discrete and continuous outcomes. Biometrics 986;42;2-30. 2. Khoury MJ, Flanders WD. Bias in using family history as a risk factor in case-control studies of disease. Epidemiology 995;6:5-9. 3. Fryer J, Silberberg JS, Wlodarczyk J, et al. Utility of two new family history measures for identifying high risk of coronary heart disease. Genet Epidemiol (in press). 4. Marenberg ME, Risch N, Berkman LF, et al. Genetic susceptibility to death from coronary heart disease in a study of twins. N Engl J Med 994;330:04-6. 5. Silberberg J. Family history of coronary heart disease: a window on disease mechanisms. Lipid Rev 994;8(2):4-8. 6. Hopper JL, Carlin JB. Familial aggregation of a disease consequent upon correlation between relatives in a risk factor measured on a continuous scale. Am J Epidemiol 992; 36: 38-47. Am J Epidemiol Vol. 47, No. 2, 998