| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
 |
ABSTRACT |
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
|
INTRODUCTION |
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
Another factor that may affect the distribution of persons affected by lymphedema is familial clustering of filarial infection. Terhell and others demonstrated that infection, as measured either by microfilaremia or antifilarial antibody status, clustered within families and was not due to shared environment.6 In a study of nuclear families in Haiti, children were 2.4–2.9 times more likely to be microfilaria positive when they had microfilaria-positive rather than microfilaria-negative mothers.7 These results suggest that filarial exposure in utero may influence the development of the fetal immune system and increase susceptibility to filarial infection. If antifilarial immunity contributes to disease development, then it is possible that lymphedema may cluster among offspring of uninfected mothers.
To address these issues, pedigrees were collected from lymphedema patients enrolled at a Haitian lymphedema treatment clinic to assess whether lymphedema prevalence was elevated in families of affected persons. We used a method proposed by Chakraborty and others to identify families with higher prevalence of lymphedema than expected from population estimates.8 The number of families exhibiting significant excess risk of lymphedema was compared with baseline population estimates of lymphedema. The data were also examined for potential bias introduced by family data collection through a clinic patient.
|
MATERIALS AND METHODS |
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
16 years of age) who lived with the patient at the time of the interview was permitted to be a proxy for the patient. The pedigree collection was administered in Creole by trained interviewers. Participants gave oral consent to pedigree collection. The study protocol was reviewed and approved by the Institutional Review Board of the Centers for Disease Control and Prevention and by the Ethics Committee of Hôpital Ste. Croix. Statistical analyses. Each patient was designated as a proband for her family. Families were sorted into two categories: families where only the proband had lymphedema and families with the proband and at least one additional relative with lymphedema. Family size, age, mean number of half-siblings, and mean number of first-degree relatives of families in these categories were compared using the Student t-test and the Wilcoxon test. Stillbirths, abortions, deaths before one year of age, adoptions, and partners of the proband who did not produce a child were excluded from analyses.
The distribution of disease in each family was compared with the expected distribution of disease obtained from census prevalence estimates of lymphedema for males and females using a permutation test.8 This test can be described as follows. Let a family of size N have members denoted by i = 1, 2, ... N. The N family members are identified through a clinic patient who is designated the proband and who is not indexed. Disease status (Xi) is determined for each family member where the ith individual may have lymphedema denoted by Xi = 1 or may not have lymphedema denoted by Xi = 0. A corresponding gender-specific population estimate of disease, pi, is assigned to each individual. Thus, the statistic
 |
depends on each family member’s disease status and an appropriate gender-specific prevalence estimate. For each set of N individuals, the observed T(X) was calculated using sex-specific prevalence estimates of lymphedema from an independent census (described later in this report). The observed T(X) was compared with an empirical distribution of T(X) values generated by permuting the data.
The analyses were repeated with the additional consideration of incomplete ascertainment. Using an approach also suggested by Chakraborty and others, the probability that an individual proband will be ascertertained (
) and family size (n) were incorporated for sensitivity analyses.8 An overall test of significance was done to assess whether the proportion of families found to have excessive lymphedema was different than expected, given the overall number of study families and available prevalence estimates. The overall test is based upon the standard test of binomial proportions.
Since the T(x) statistic follows a binomial distribution made of discrete units, we anticipated some difficulty identifying a statistical cut-off value that would match the testing level of 
= 0.05 or equivalently, a cumulative probability equal to 95%. This occurred because T(x) follows a binomial distribution defined by N family members and the proportion of family members with disease. The probability distribution for the range of possible T(X) values has gaps which correspond to gaps between the count data, i.e., it was not possible for 3.4 relatives to have disease. For some families, the observed T(X) distribution corresponded to a cumulative probability greater than 95%, while the cut-off T(X) value corresponding to 95% cumulative probability was not available. Instead, the next closest value to the observed T(X) corresponded to a cumulative probability less than 95%. We distinguished between families with a T(X) cut-off value with a true correspondence to 95% cumulative probability and families with a T(X) cut-off value corresponding to a cumulative probability less than 95%. We calculated cumulative probabilities for both groups, but retained only the group that had a T(X) value with true correspondence to the desired testing level.
Choice of ascertainment probabilities .
The initial statistic assumes that sampling of the family is through a totally random process. In fact, families are often enrolled by identifying a single proband, which is not a random process. Thus, we followed the standard practice of excluding the proband from the analysis. Since we disrupt the random sampling, a statistical adjustment for the probability of an affected individual becoming a proband () addressing the alteration in sampling procedure is provided.8
The families were evaluated at = 0.5, which reflected earlier observations that the clinic contained approximately 50% of the lymphedema cases living in the clinic catchment area.9 As a sensitivity analysis, the families were also evaluated under the assumption of a lower probability of lymphedema cases in the population being enrolled at the clinic, i.e., = 0.1.
Estimates of lymphedema prevalence. Estimates of lymphedema prevalence among men and women were obtained from a census conducted between 1990 and 1991 in a nearby Haitian community.10 The prevalence of lymphedema among men and women who were at least 16 years old was 0.002 (1 of 439) and 0.024 (14 of 588), respectively. These disease estimates concur with the model-based prevalence estimate (0.02285), which combines disease and infection prevalence in generating an estimate for Haiti.11 The model estimate is based on infection estimates and imputed disease levels from population estimates of countries other than Haiti. We chose to use the empirically derived estimates from the Haitian community rather than the estimates from theoretical modeling.
Although the population prevalence estimates are from a community census, the statistic used in the analysis is sensitive to sex-specific prevalence. We assessed the sensitivity of the test statistic to variability in the gender-specific prevalence estimates. Three types of families were observed in the data with patterns of secondary cases of disease distributed differently by sex: 1) one female with disease, zero affected males, 2) one male with disease, zero affected females, and 3) one male and one female with disease. The statistics for families within a single category are expected to respond to changes in the assumptions of prevalence similarly. One family was selected from each category for the sensitivity analysis; the results from those three families are generalizable to the remainder of families within the respective categories. Using three representative families, we compared two male:female prevalence sets with the census estimates of 0.002 and 0.024 for males and females, respectively. Sensitivity analyses were selected assuming that the female estimate might be too high and the male estimate too low from the census. First, the female prevalence estimate was fixed. Since the male estimate was unstable because it was determined by a single event, we hyperinflated the estimate to 0.01. The other prevalence combination addressed in the sensitivity analysis assumed that the male census estimate was accurate. It remained fixed while the female estimate, which was based on more events in the census, was decreased to 0.020.
|
RESULTS |
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
). There were 53 additional cases of lymphedema, 12 males and 41 females, found in those families. While multiple affected families were slightly larger in family size, the difference was not significant at a 0.05 level. The mean number of half-siblings was less for multiple affected than for single affected families (P = 0.058). The mean number of first-degree relatives was higher in multiple affected families, but not significantly so.
|
). Of these 15 families with excess disease, 12 families had an affected male. The remaining families with excess disease had either two females with lymphedema or very few female family members. An additional 28 families also were detected to have excess disease based upon family-specific cumulative probabilities. While these families are likely to have excess disease, a conservative approach was taken by considering that the group of 28 families did not have excess disease because the statistical cut-off point did not correspond precisely with the cumulative probability of 95% (data not shown).
|
). The prevalence of lymphedema reported in relatives of all probands was 0.023 (53 of 2,256). The families with excess lymphedema had slightly fewer females and more males than families without excess lymphedema. There was no significant difference between the mean number of half-sibs and first-degree relatives per family between categories of families.
|
= 0.5 was incorporated into calculations, T(X) values generally increased and cumulative probability increased slightly (Table 2
). Three families, in addition to the 15 families identified by the Chakraborty test, were identified as having cumulative probabilities 0.95 (data not shown). These three families had either two affected females or relatively few female family members. When an ascertainment correction was applied using = 0.1, the observed T(X) values were identical to those from = 0.5, but the cumulative probability tended to fall between values generated from the Chakraborty test uncorrected for ascertainment and = 0.5 scenarios (Table 2). Overall, the correction for ascertainment through a female proband using 3 = 0.1 and 0.5 resulted in larger cumulative probabilities (further in the tail of the distribution) than the Chakraborty test, which translated to greater statistical significance being observed. In other words, the initial uncorrected T(x) values were conservative and were not altered by correcting the analyses for ascertainment through a single proband.
Influence of prevalence estimates.
Using the Chakraborty test with pf = 0.024 and pm = 0.002, the number of families exhibiting excess disease (n = 15) was significantly greater than the number of families expected to have excess disease out of the 172 families (P = 0.026). When = 0.1, there was an additional family identified with excess disease and the overall P value was 0.013. When the three additional families found to have excess disease when = 0.5 were added to the original 15 families with excess disease, the overall P value was calculated to be 0.0023. When lymphedema prevalence was lowered for females to pf = 0.020, the observed T(X) values increased and resulted in three additional families having evidence of excess disease (data not shown).
Additional sensitivity analyses.
Within the three families representative of male-female disease patterns, increasing lymphedema prevalence for males from pm = 0.002 to pm = 0.01 led to a decrease in observed T(X) values for families 3 and 16 (Table 4).
|
0.05, 14 female-only families with excess disease would be needed. |
DISCUSSION |
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
The results from this study were consistent with the conclusion that lymphedema of the leg clusters in some families. Demographic differences between families with multiple cases and single cases of lymphedema do not explain the findings. Analyses to detect whether any bias was introduced by sampling families non-randomly reveal that sampling through a proband had a negligible impact on significance testing. Statistical attributes, missing data, and sensitivity issues need to be considered when interpreting these data. These are the first families that have been identified that appear to share some common exposure, either genetic or environmental, linked to the risk of developing lymphedema due to lymphatic filariasis.
Ascertainment through a single proband does not appear to have a large impact on statistical results. The unadjusted Chakraborty test and the adjusted approach accounting for potential ascertainment bias give similar though not identical results in terms of the number of families identified as potentially exhibiting excess risk.
Twenty-eight families were categorized as not meeting significance criteria. By erring on the conservative side and treating those families as not having excess disease, the number of families with excess disease was undercounted, leading to more conservative overall testing for significance across families.
These findings need to be interpreted with caution because of the possible effect that missing age data and the gender-specific prevalence estimates may have on the analyses and statistical testing. While the presence of excess disease appears more convincing for families with diseased males, it must be kept in mind that we have overestimated the size of the at-risk population. Based on reported age of onset for filarial lymphedema, family members should have reached adolescence, at a minimum, to be considered eligible for inclusion in the population at risk for lymphedema. The number of persons at risk for the disease outcome is likely to be inflated in the study population since ages were unknown for most of the youngest generation. If these young persons would have been excluded from analyses, it would have been easier to attain statistical significance. It is possible that some of this generation was at risk for disease so they were retained for analyses. As a consequence of including persons of unknown age in the analyses, the estimated proportion of diseased persons may be smaller than the true value. The inclusion of extra persons in the denominator may result in overly conservative testing.
The analyses presented here are sensitive to initial prevalence estimates, especially for males. Better prevalence estimates stratified by gender and other demographic characteristics might alter these results. Ongoing census data collection in additional communities in the area might provide more detailed prevalence estimates that could be stratified on demographic characteristics. Our results suggest that census estimates of disease prevalence by gender used for initial analyses were reasonable, albeit more so for females. These estimates were obtained from a community-wide household sample. All household members were assessed for disease during the process. The estimated prevalence of disease in the study probands’ relatives without regard to age was similar to that found in census participants more than 16 years old. Statistical testing indicated that the expected infrequent occurrence of lymphedema in a male family member led to a family with a male lymphedema case being designated as having excess disease. By artificially increasing the estimates of the baseline population prevalence of lymphedema in males during statistical testing, the effect of an affected male on the statistical test was diminished. In other words, the presence of a male lymphedema case becomes less unusual and this influences the outcome of the clustering analyses. These analyses suggest that additional work is needed to establish reliable estimates of lymphedema prevalence, especially for males. If the estimates used here are accurate, the presence of males in families noted to have excess disease may just be a reflection of the disease aggregation. The detection of affected males may flag families with higher risk of disease.
While the results suggest that familial aggregation of lymphedema is present in some families, these findings alone do not indicate whether host genes or environment regulate pathogenesis. However, there is some suggestive evidence that antifilarial immune responses from a subset of the families are influenced by genetics even after controlling for environmental factors (T.Cuenco K, unpublished data). If immune responses are risk factors for disease development and are controlled by genes, then lymphedema pathogenesis would be expected to have a genetic basis. If a portion of the aggregation is due to genetic factors, this would lend support to the idea that the immune response influences lymphedema development and that shared human leukocyte antigens (HLA) or other immune response genes that exhibit variation in the human population may differ among affected persons. The evidence is mixed for this hypothesis. Specific HLA alleles, which are associated with a number of other diseases, have been associated with elephantiasis, a severe form of lymphedema.12 Patients were more likely to be HLA DR3 negative (relative risk [RR] = 0.25, 95% confidence interval [CI] = 0.067, 0.27) and DQ5 positive and (RR= 3.55, 95% CI = 1.19, 10.56). The absence of DR3 and 2B3 was more frequent in patients versus controls (RR = 0.25 and 0.31, respectively). Conversely, an earlier study found combinations of HLA–A and –B gene products not to be associated with disease or infection.13 While these results contradict the later study, it is possible that they reflect differences in methods or study populations. Assuming that lymphedema is influenced by host immune responses, these results indicate that there might be a genetic basis to lymphedema pathogenesis.
If we assume that there is a genetic component to lymphedema development, then it could prove useful to determine which genes are shared by relatives with disease. The information may provide insight into possible gene-specific interventions. Disease development also may be influenced by proximity to vector habitat or vector exposure. Families with excess disease can then be assessed by further environmental testing while controlling for the degree of genetic sharing.
This is the first family study based on pedigrees to assess familial aggregation of lymphedema due to lymphatic filariasis. The presence of excess disease in certain families suggests that members of those families might be prone to developing lymphedema after infection by W. bancrofti. This implies that an underlying genetic or environmental process common to individuals from a particular family might be responsible for the observed disease distribution. Additional studies of these families could lead to a better understanding of pathogenesis and development of new therapeutic strategies. If the excess disease found in a family was the result of a shared environmental factor, rather than a genetic factor, then prevention efforts could be targeted at altering the environmental factor.
Received March 23, 2003. Accepted for publication November 6, 2003.
Acknowledgments: We extend special thanks to Jacky Louis-Charles, Antoine Florent Michelus, Carline Casseus, Yvrose Dumond, and Erick Leriche for their tireless efforts in interviewing community members; David Addiss and the staff of the Hôpital Ste. Croix and its Lymphedema Clinic for their assistance with the project; and Amanda R. Freeman and John J. Hanfelt for their assistance. We also thank Centers for Disease Control and Disease Prevention Genetics Working Group for providing funding that was essential for conducting the study.
Authors’ addresses: Karen T.Cuenco, Department of Epidemiology and Preventive Medicine, University of Maryland School of Medicine, 660 West Redwood Street, Baltimore, MD 21201, Telephone: 410-706-2229, Fax: 410-706-04425. M. Elizabeth Halloran, Department of Biostatistics, Rollins School of Public Health, Emory University, 1518 Clifton Road NE, Atlanta, GA 30322, Telephone: 404-727-7647, Fax: 404-727-1370. Patrick J. Lammie, Division of Parasitic Diseases, National Center for Infectious Disease, Centers for Disease Control and Prevention, Mailstop F-13, 4770 Buford Highway Atlanta, GA 30341, Telephone: 770-488-4054, Fax: 770-488-4108, E-mail: pjl1{at}cdc.gov.
|
REFERENCES |
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
