Validity of family history data on PD

Population-based cases and controls.

We used the medical records-linkage system of the Rochester Epidemiology Project to identify all subjects residing in Olmsted County, MN, who developed PD from 1976 through 1995.10,11⇓ Details about the study population and the identification of incident cases were reported elsewhere.11 Our diagnostic criteria included two steps: the definition of parkinsonism as a syndrome and the definition of PD within the syndrome. Parkinsonism was defined as the presence of at least two of four cardinal signs: rest tremor, bradykinesia, rigidity, and impaired postural reflexes. PD was defined as the presence of parkinsonism with all three of the following: no secondary cause (e.g., repeated stroke with step-wise progression, repeated head injury, history of encephalitis, neuroleptic treatment within 6 months before onset, hydrocephalus, brain tumor); no documentation of unresponsiveness to levodopa at doses of at least 1 g/d in combination with carbidopa (applicable only to patients who were treated); and no prominent or early (within 1 year of onset) signs of more extensive nervous system involvement (e.g., dementia or dysautonomia) not explained otherwise.11 Our clinical classification of patients with PD through medical records review was found to be valid compared with a direct examination by a movement disorders specialist, as reported elsewhere.12

Each PD case subject was individually matched by age (±1 year) and sex to a general population control subject residing in Olmsted County, MN, and free of PD, other parkinsonism, or tremor of any type in the year of onset of PD in the matched case (index year). Further details about the selection of controls were reported elsewhere.12

Family history method.

Whenever case or control subjects were alive and able to undergo a telephone interview, a questionnaire was administered to them including a detailed family composition section. For case and control subjects who died before the study, were incapacitated, or elected a proxy interview, we identified an initial contact person. Through a telephone call with this initial contact person, we then identified the most knowledgeable person to serve as the informant for the family history method (“best proxy”).

We asked case and control subjects (or their best proxies) about the name, vital status, and other demographic information of parents, siblings, and children (of the case and control subjects). After the full listing of these first-degree relatives was obtained, we asked the question, “Has anyone in your (his or her) family had PD?” If the case and control subjects (or their best proxies) responded positively, follow-up questions addressed the name, relationship, and sex of the affected relatives. We excluded from this validity study the complete family of case and control subjects who refused to participate in the interview or those for whom a proxy respondent could not be identified.

Family study method.

The primary methodology of the family study involved a two-phase design with individual screening of the relatives for PD and examination of those screened positive (figure). Following the interview described previously, we mailed to case and control subjects (or their best proxy) a family data form to obtain additional information for each of the first-degree relatives including their address, telephone number, and permission to contact. A screening instrument for PD was administered to living first-degree relatives through a direct telephone interview.13 For relatives who died before the study, were incapacitated, or elected a proxy interview, we administered the screening instrument to a proxy informant. Relatives screened positive if they had a previous diagnosis of PD, a previous treatment with levodopa, at least two of nine symptoms of parkinsonism, or “shaking” alone.13 Living relatives who screened positive were invited for an examination by a movement disorders specialist either at the Mayo Clinic or at home. For relatives who screened positive but could not be examined and for those who screened positive but were deceased, we obtained copies of their medical records (from institutions or physicians not included in the records-linkage system).

• Download figure
• Open in new tab
• Download powerpoint

Figure. Flow charts indicating the steps involved in the family study method for relatives of case subjects (left) and relatives of control subjects (right). *A total of 378 relatives were screened dependently (i.e., through the case subjects or their best proxy) and 30 were not screened (378 + 30 = 408). Of these 408 relatives, information on PD status independent of the screening was obtained from examination for 17 or from medical record for 134 relatives. These 151 relatives with independent documentation of PD status were included in the study. In total, 655 relatives were included in the study (504 screened independently plus 151 not screened independently). †Despite the negative screening, examination was conducted on 52 relatives and an adequate medical record was available for 94 relatives. These sources allowed the detection of two additional cases of PD (among those who screened negative). ‡A total of 393 relatives were screened dependently (i.e., through the control subjects or their best proxy) and 22 were not screened (393 + 22 = 415). Of these 415 relatives, information on PD status independent of the screening was obtained from examination for 17 or from medical record for 122 relatives. These 139 relatives with independent documentation of PD status were included in the study. In total, 511 relatives were included in the study (372 screened independently plus 139 not screened independently). §Despite the negative screening, examination was conducted on 20 relatives and an adequate medical record was available for 61 relatives. These sources did not allow the detection of any additional cases of PD (among those who screened negative).

Independent of the screening method outlined here, two strategies provided additional information on the PD status of particular subgroups of relatives. First, all relatives residing within 120 miles of Rochester, MN, and aged ≥60 years were invited for examination (at Mayo or at home) regardless of the screening. Second, all relatives who resided in Olmsted County, MN, were studied through the records-linkage system regardless of the screening.10 For the relatives who could not be screened, or who screened negative, information in the records-linkage system was considered adequate if a relative had been in contact with the system at least once in the year preceding the time of the study (or preceding death). Whenever relatives who screened positive refused the examination and medical records could not be obtained (from within or outside the records-linkage system), we accepted the verbal report of a PD diagnosis. For relatives with multiple sources of information, the final PD status was based on the best available information in hierarchical order (examination, medical records, or interview). To avoid circularity between the family history method and the family study method, relatives for whom information was obtained only through the interview of case and control subjects (or of their best proxies) were excluded. In addition, those first-degree relatives who served as best proxies were excluded.

Data analysis.

We estimated sensitivity and specificity of the family history method using the family study method as the standard for comparison. Sensitivity was the ability of case and control subjects (or their best proxies) to report as affected the relatives found to have PD by the family study method (true positives divided by true positives plus false negatives). Specificity was the ability of case and control subjects (or their best proxies) to exclude PD in relatives found to be free of PD by the family study method (true negatives divided by true negatives plus false positives).14 We also computed exact binomial confidence intervals for sensitivity and specificity. We conducted overall analyses and analyses stratified by characteristics of the case and control subjects (or their best proxies) and by characteristics of the relatives. Statistical testing of differences in sensitivity and specificity across case and control subjects or across strata (e.g., direct vs proxy interview for cases) was carried out using generalized estimating equations with a log link and an exchangeable correlation structure, to account for the clustering of relatives within families.15-17⇓⇓

To estimate the magnitude of the bias introduced by the family history method, we compared case-control results from the family history method with results from the family study method. Positive family history was defined as having at least one first-degree relative with PD. The original matching of case and control subjects was not taken into account because of the restriction of the sample to individuals who could be studied independently using the family history and the family study methods. The odds ratio (OR) and 95% CI were computed using logistic regression models adjusted by the sex of case and control subjects, sibship size (in quartiles), and age at onset of PD or index year (in quartiles).18 The magnitude of the bias was measured using the relative bias (OR from the family history method minus OR from the family study method, divided by OR from the family study method).19 Statistical testing was done at the conventional two-tailed α level of 0.05. All analyses were performed using the SAS package (Cary, NC).20

Results.

Our findings show that case subjects (or their best proxies) have higher sensitivity than control subjects (or their best proxies); i.e., they are more aware of other members of the family affected by PD. However, case subjects (or their best proxies) overreport PD in relatives who are not affected (lower specificity).21 The difference in specificity may seem at first negligible (1%); however, it was in part responsible for the family information bias that we observed. The almost 100% specificity in controls (99.8%) resulted from only one relative incorrectly labeled with PD among 500 relatives free of PD (by the standard). By contrast, the almost 99% specificity in population-based cases (98.7%) resulted from eight relatives incorrectly labeled with PD among 630 relatives free of PD (by the standard). A 1% difference in specificity caused a substantial difference in absolute numbers (1 vs 8 probands misclassified as positive family history) because the numbers of unaffected relatives were large. Thus, both the lower sensitivity in control subjects and the lower specificity in case subjects contributed to the bias.

Four of the eight relatives incorrectly reported to have PD by case subjects (or their best proxies) were affected by a type of parkinsonism other than PD and one additional relative had essential tremor. The single relative incorrectly reported to have PD by control subjects (or their best proxies) had a type of parkinsonism other than PD. The risk of misclassifying relatives affected by another type of parkinsonism or by essential tremor as PD is substantial and should be considered when interpreting findings from the family history method.

In a distinct but related study, we investigated the validity of the family history method among a sample of 368 patients with PD (220 men and 148 women) referred to the Department of Neurology of the Mayo Clinic in Rochester, MN, from July 1996 through October 2000 (referral cases). Their median age at onset of PD was 62 years (range = 28 to 85), their median duration of PD was 3 years (range = 0 to 27), and their median age at the time of the study was 68 years (range = 31 to 90). Almost all of the case subjects were directly interviewed (98%), and they yielded 1,759 first-degree relatives. Thirty-three of the relatives (1.9%) had PD according to the family study method; the sensitivity of the family history method was 70% (95% CI = 51 to 84) and its specificity 99% (95% CI = 98 to 99) compared with the family study method (data not shown in tables). Even though these referral case subjects were younger and more often directly interviewed than the population-based case subjects, estimates of sensitivity and specificity were similar in the two groups.

We did not find any significant determinant of sensitivity; in particular, there were no significant differences in sensitivity between direct interviews and proxy interviews. This is in part caused by the small number of PD cases that were detected among the relatives of case and control subjects, despite the overall large size of the study. Although the differences in sensitivity were not significant, combined with differences in specificity they determined the variability in relative bias across strata. Our results suggest that the relative bias was more pronounced for proxy interviews, for women serving as informants, when the relatives were siblings, when the relatives were alive at the time of the study, and when the relatives were examined or had medical record documentation (see table 3, last column to the right).

All previous studies on the familial aggregation of PD have been based on the family history of PD as reported by case and control subjects (or one proxy informant per family).1-8⇓⇓⇓⇓⇓⇓⇓ Our findings suggest that the OR may have been overestimated; however, the extent of the bias in previous studies cannot be quantified because the validity of the family history method is likely to vary across the studies. The phrasing of the questions used to collect information from case and control subjects (or their proxies) varied across studies. For example, we used a single question about PD following a detailed listing of each relative (including name, vital status, and other demographic information). Alternative approaches range from a minimum of asking a single question about the family as a whole to a maximum of repeating the same question for each family member separately. Sensitivity may also be increased by asking additional questions about symptoms of parkinsonism; however, in the absence of a follow-up examination or a review of medical records, the specificity may be reduced substantially.

There are only two previous small-scale studies of the validity of the family history method for PD; neither of them involved a control group. In one study conducted in New York, NY, the investigators reported 100% sensitivity in five relatives with PD and 100% specificity in 56 relatives without PD.22 In a Canadian study, patients underreported parkinsonism in their family; however, no sensitivity and specificity estimates were provided.5

Although OR estimates from the family history method were reported in table 3, these estimates were restricted to the subsample of case and control subjects that could be studied independently through the family history and the family study methods (62% of the total sample). The primary results of our study on the familial aggregation of PD, including the complete sample and using a historical cohort design (survival analyses accounting for age of relatives and censoring) will be reported elsewhere (Rocca, unpublished findings).

Methods.

Our study has a number of strengths. First, our validation study was relatively large. A study of this magnitude was only possible in the context of a study of the familial aggregation of PD that provided the justification and the funding for the direct contact of a large number of first-degree relatives. Second, our case and control subjects were derived from a well-defined population that is served by a unique records-linkage system.10 This system provided medical record information for a large proportion of the relatives (see figure).

Our study also has limitations. First, the family history method used in this study was imperfect. Case and control subjects (or their best proxies) were asked if any of their relatives had PD. If they gave a positive answer to this question, they were further asked to indicate the names of the affected relatives and their relation; however, the question was not repeated for each relative. We considered unaffected all the relatives that the case and control subjects (or their best proxies) did not identify as having PD. Unfortunately, the negative responses are likely to include a proportion of unknown responses. However, sensitivity analyses show that only a large difference in the proportion of unknown responses between case and control subjects would affect our findings (data not shown).

Second, our estimates of sensitivity and specificity were obtained after excluding 69 case subjects (34%) and 83 control subjects (41%) who refused to participate or could not be studied independently by the family history and the family study methods. If case and control subjects who were excluded (or their best proxies) were more or less aware of their family history than those who were included, our estimates of sensitivity and specificity could be biased.21 Case subjects who were included were younger at onset of PD and at the time of the study than the case subjects excluded; however, they were not different by sex, vital status, and duration of PD. By contrast, control subjects who were included were similar to those excluded for all the variables considered. It remains uncertain whether the selective participation of younger case subjects with PD may have influenced our findings. All our case-control analyses for family history of PD reported in table 3 were adjusted by age at onset of PD (or index year for controls) in addition to sex (of case and control subjects) and sibship size.

Third, because case and control subjects were recruited over a 20-year period, a number of them were deceased at the time of the study and some were severely incapacitated. As a result, 64% of the case subjects and 50% of the control subjects could only be studied through a proxy (best proxy). The involvement of best proxies increased the complexity of our study design. However, the sensitivity and specificity of the family history method did not vary greatly between direct and proxy interviews. We also performed stratified analyses for individuals with incidence of PD (or index year) from 1976 through 1985 and from 1986 through 1995. The relative bias was more pronounced for the more recent decade (297%) than for the previous decade (35%). The major reason for this difference was the percentage of relatives who were alive at the time of the study. Relatives who were alive at the time of the study could be screened directly and examined if they screened positive; therefore, the quality of the standard for comparison (the family study method) was higher. For the early decade, the difference between the family history method and the family study method was reduced by the inability to directly study a large segment of relatives.

Fourth, the quality of the standard varied across relatives. Because the family study method involved three strategies, a two-phase screening and examination design, the examination of a number of relatives regardless of the screening, and the review of medical information in the records-linkage system, PD status was derived for each relative from the best available documentation. Therefore, the best available documentation was a direct examination for a segment of relatives, a medical record for another segment, a direct interview for another segment, and a proxy interview for the remainder (see figure). The relative bias was more pronounced when the standard included a neurologic examination or medical record documentation compared with telephone interview only. In the absence of a direct examination or of medical record documentation of the PD status among relatives, the family history method and the family study method yielded similar results.

Finally, in the two-phase screening and examination approach, the relatives were invited for examination if they screened positive at the telephone interview. Therefore, the validity of the family study method relied largely on the validity of our telephone screening instrument for PD. We have previously reported that the sensitivity of the telephone screening instrument was high13; therefore, the number of false negatives should be negligible.