A Prospective 4- to 5-Year Study of DSM-III-R Hypochondriasis

Background  Although hypochondriasis is generally thought to be a chronic and stable condition with a relatively low remission rate, this disorder remains understudied.

Methods  This is a 4- to 5-year prospective case-control study of DSM-III-R hypochondriasis. Medical outpatients meeting DSM diagnostic criteria for hypochondriasis completed an extensive research battery assessing hypochondriacal symptoms, medical and psychiatric comorbidity, functional status and role impairment, and medical care. A comparison group of nonhypochondriacal patients from the same setting underwent the same battery. Four to 5 years later, both cohorts were re-interviewed.

Results  One hundred twenty hypochondriacal and 133 nonhypochondriacal comparison patients were originally studied. Follow-up was obtained on 73.5% (n=186) of all patients. At follow-up, the hypochondriacal sample was significantly (P<.001) less hypochondriacal and had less somatization (P<.001) and disability than at inception, but 63.5% (n=54) still met DSM-III-R diagnostic criteria. When compared with the comparison group using repeated measures multivariate analysis of variance, these changes remained statistically significant (P<.0001). Changes in medical and psychiatric comorbidity did not differ between the 2 groups. When hypochondriacal patients who did and did not meet diagnostic criteria at follow-up were compared, the latter had significantly less disease conviction (P<.05) and somatization (P<.01) at inception, and their incidence of major medical illness during the follow-up period was significantly (P<.05) greater.

Conclusions  Hypochondriacal patients show a considerable decline in symptoms and improvement in role functioning over 4 to 5 years but two thirds of them still meet diagnostic criteria. Hypochondriasis, therefore, carries a very substantial, long-term burden of morbidity, functional impairment, and personal distress.

LITTLE IS known about the natural history of hypochondriasis, although it is generally thought to be a chronic and stable condition with a relatively low spontaneous remission rate.^1-10 More favorable outcomes are suspected in those without personality disorder, comorbid Axis I disorder, or accompanying stressful life events, and in those whose hypochondriasis is of shorter duration and accompanied by serious medical comorbidity.^1-10 The empirical evidence for these impressions, however, is sparse and inconclusive. Recent studies have supplemented our understanding. Noyes et al¹¹ conducted a prospective, case-controlled, follow-up of 48, DSM-III-R hypochondriacal patients. After 1 year, 67% (n=32) of the hypochondriacal patients still met diagnostic criteria for hypochondriasis, and these patients were significantly higher than the comparison group on measures of hypochondriacal symptoms, somatization, anxiety and depressive symptoms, and medical utilization. Improvement in hypochondriasis was predicted by a shorter duration of hypochondriasis at inception and fewer symptoms of depression and neuroticism. Robbins and Kirmayer¹² conducted a prospective, 1-year follow-up of patients with hypochondriacal responses on an illness worry questionnaire. Hypochondriacal concerns declined significantly in one half, but they remained more concerned about their health at follow-up than nonhypochondriacal patients. Hypochondriasis was persistent in those with more medical comorbidity and more depressive and anxiety disorders.

We conducted a 4- to 5-year, prospective, case-control, cohort study of DSM-III-R hypochondriasis. At inception, patients underwent a criterion standard, diagnostic interview, and assessment of hypochondriacal symptoms, medical and psychiatric comorbidity, functional status, and medical care.^13-19 A comparison group of nonhypochondriacal patients from the same medical setting underwent the same battery. In this article we describe and compare the clinical characteristics of the 2 groups at follow-up, and we examine the changes occurring over the follow-up interval.

Subjects were recruited in 1987-1989 (cohort 1) and 1990-1992 (cohort 2) by screening consecutive patients attending a primary care clinic on randomly selected days with a self-report hypochondriasis questionnaire. Those exceeding a predetermined cutoff, and a random sample of patients below the cutoff, returned at a later date to undergo the research battery. This included a criterion standard, diagnostic interview, and those diagnosed as hypochondriacal on this interview constituted the hypochondriacal sample. After the research interview, the medical record was audited to determine aggregate medical morbidity. Although each individual patient completed the same research battery at inception and follow-up, the batteries administered to cohorts 1 and 2 contained most, but not all, of the same instruments.

Four to 5 years later, all patients were asked for their informed consent to repeat the research battery. This was administered by trained interviewers, blinded to the inception data. Because of disability or geographic distance, several of the follow-up assessments were conducted by telephone. Medical records were then audited for medical morbidity by a study physician (E.D.B.) blind to the other research data.

The study was conducted in a large primary care clinic of an academic teaching hospital. Cohorts 1 and 2 had slightly different eligibility criteria: cohort 1 patients had been in treatment at the hospital for at least 24 months and had visited their primary care physician at least once before, while cohort 2 patients had no such inclusion criteria; the patients of house officers were excluded from cohort 1 but not from cohort 2.

Hypochondriacal symptoms were assessed with the Whiteley Index^20-24 and the Somatic Symptom Inventory (SSI).^25,26 The Whiteley Index consists of 14 hypochondriacal attitudes and concerns, scored on an ordinal scale from 1 to 5,^20-24 and contains 3 factors: disease fear, disease conviction, and bodily preoccupation. Its test-retest reliability and discriminant and convergent validity have been established.^20-24 The intrascale consistency of the Whiteley Index in a previous study was 0.85 (Cronbach α), and the test-retest reliability was 0.84 (Pearson's product-moment correlation) over a mean interval of 25.6 days.¹⁷

Hypochondriacal somatic complaints (somatization) were assessed with a 26-item SSI.^25,26 Responses are scored on an ordinal scale from 1 to 5. In previous work,¹⁹ this scale had a test-retest reliability of 0.86 (Pearson product-moment correlation) over a mean interval of 25.6 days and an intrascale consistency of 0.95 (Cronbach α). The Whiteley Index and SSI were intercorrelated (R=0.52; P<.001) and together identified medical outpatients with a syndrome consistent with DSM-III hypochondriasis.¹⁹

The clinical diagnosis of hypochondriasis was made with the Structured Diagnostic Interview for Hypochondriasis based on operationalized DSM-III-R criteria.²⁷ Interrater agreement with this instrument is 96%, and the univariate correlation between the interview responses and self-report questionnaire scores is 0.75.²⁷ The DSM diagnosis of hypochondriasis specifically excludes hypochondriacal symptoms that are better explained by another, comorbid psychiatric disorder or by major medical illness. Axis I Psychiatric Disorder. Psychiatric disorder was assessed with the Diagnostic Interview Schedule (DIS).²⁸ Cohort 1 received version 3-A while cohort 2 received version 3-R. This widely used, highly structured interview provides diagnoses in conformity with DSM criteria. Its validity, reliability, and psychometric characteristics have been extensively examined.^28,29 The sections covering psychosexual dysfunction, posttraumatic stress syndrome, and tobacco dependence were omitted. The hierarchical exclusion criteria of DSM-III were not employed.

Personality disorder "caseness" was assessed with the 5-item impairment/distress subscale of the Personality Diagnostic Questionnaire–Revised.^30,31 These 5 dichotomous items correlate 0.75 with the total score on the complete Personality Diagnostic Questionnaire and can be used to predict caseness, ie, the likelihood of any DSM-III-R personality disorder being present.³² Thus, we can only report the likelihood of the patient's having a personality disorder but cannot specify which particular disorders might be present. A score of 2 or higher has been used as an indicator of personality disorder caseness.³³

The Somatosensory Amplification Scale^34,35 is a 10-item questionnaire assessing self-reported sensitivity to normal physiologic states and minor bodily sensations that are not generally regarded as symptomatic of serious disease.^34,35 The Somatosensory Amplification Scale has a test-retest reliability of 0.79 over a median interval of 74 days and an internal consistency of .82 (Cronbach α). It is significantly associated with DSM-III-R hypochondriasis, even after controlling for psychiatric comorbidity,³⁵ and it predicts the persistence of hypochondriacal symptoms in transiently hypochondriacal patients.¹⁷

Disability was assessed with the Functional Status Questionnaire, a self-report instrument designed for use with ambulatory medical patients, whose intrascale reliability and construct validity have been demonstrated.³⁶ We used the 31 items composing 4 of its 6 subscales: basic activities of daily living, intermediate activities, social activities, and work performance. The Functional Status Questionnaire also asks the number of days of curtailed activity and of days spent in bed because of illness in the past month. The Functional Status Questionnaire scores range from 0 to 100 (maximal function).

Aggregate medical morbidity was assessed from the medical record. All medical diagnoses were rated for seriousness with explicit, preestablished criteria (including extent of tissue damage, clinical progression, complications, recurrence, and threat to life).³⁷ Each diagnosis was assigned a weight of 1 through 4 and then summed as an index of aggregate morbidity (1 indicates major; 2, moderate; 3, minor; and 4, nonspecific complaint). Examples of major diagnoses included myocardial infarction, lung cancer, and renal failure; examples of moderate diagnoses included asthma, hepatitis, and gout. In previous work, this method had moderate interrater reliability and correlated 0.41 (P<.001) with primary physician ratings of morbidity.³⁷

The data analytic strategy was to employ multivariate techniques where appropriate to accommodate the multicolinearity among variables and to test for differences between groups over time with domains of variables. Data analysis was conducted in phases. First, descriptive statistics and reliability coefficients were calculated for each cohort separately, and significant differences between and within hypochondriacal and comparison groups were determined by independent or paired Student t tests, or χ² test. Alpha was set at <.05 for all analyses. Second, preliminary analyses were performed to assess differences between cohorts by independent Student t tests and χ² as appropriate. The 2 cohorts were then combined for all subsequent analyses. Third, to test for changes over time, repeated measures multivariate analysis of variance (MANOVA) was performed, with groups (hypochondriacal vs comparison patients) as the between-subjects factor and time (inception and follow-up) as the within-subjects factor. Age and education were included as covariates in the models. A similar MANOVA with repeated measures approach was used to analyze differences between patients who did (nonremitters) and did not meet (remitters) diagnostic criteria 4 to 5 years later. This approach takes into account the group differences at inception and changes occurring during the follow-up interval and permits an overall test of significance, accommodating the expected multicolinearity among the dependent measures. Assuming statistical significance for the multivariate test, then separate univariate, repeated measures, analysis of variance (ANOVA) tests of significance for each dependent measure were calculated, adjusting for inflated F values using the Greenhouse-Geiser method.

Changes in DSM nosology and in the DIS interview over the course of the study necessitated the development of an index of aggregate psychiatric morbidity that was comparable at inception and follow-up. An index was constructed consisting of the total number of diagnoses present, including major depression, dysthymia, panic disorder, phobias, obsessive-compulsive disorder, and alcohol abuse.

Cohort 1 was composed of 60 DSM-III-R hypochondriacal patients and 73 comparison patients, from among 2889 patients screened. At follow-up, 3 (5%) of the hypochondriacal patients could not be recontacted, 2 (3.3%) had died, 4 (6.7%) were unable to participate (eg, severe medical illness, senility, aphasia), and 4 (6.7%) refused. Therefore, 47 patients (87% of those eligible and 78.3% of all subjects) were reinterviewed (5 by telephone). In the cohort 1 comparison sample, 3 subjects (4.1%) could not be contacted, 6 (8.2%) had died, 1 (1.4%) was unable to participate, and 4 (5.5%) refused. Thus, 59 (89.4% of those eligible, and 80.8% of all subjects) were reinterviewed (3 by telephone). The mean (±SD) follow-up interval was 64.7 months (±6.8 months) for the hypochondriacal sample and 64.1 months (±6.4 months) for the comparison group.

Cohort 2 was composed of 60 DSM-III-R hypochondriacal patients and 60 comparison patients from among 2389 patients screened. In the hypochondriacal sample, 2 patients (3.3%) had died, 6 (10%) could not be contacted, and 14 (23.3%) declined to participate. Thus, 38 patients (65.5% of those eligible and 63.3% of the total inception cohort) were reinterviewed (2 by telephone). In the cohort 2 comparison sample, 4 patients (6.7%) had died, 2 (3.3%) could not be contacted, and 12 (20%) declined to participate. Thus, 42 patients (75% of those eligible and 70% of the total cohort) were reinterviewed (5 by telephone). The mean (±SD) follow-up interval was 50.2 months (±5.0 months) for the hypochondriacal sample and 51.3 months (±5.6 months) for the comparison group.

We first compared the inception data of the 2 cohorts and of those patients who were and were not followed-up. Patients in cohort 1 were older than those in cohort 2: 54.5 years (±15.0 years) vs 49.0 years (±16.1 years) (P<.02), but they did not differ significantly on sex, marital status, education or social position, or race. Cohort 1 patients were medically sicker than cohort 2 patients at inception, having a mean of 0.93 major medical diagnoses (±1.43) in their medical records as compared with 0.42 (±0.91) for thos in cohort 2 (P<.001). A significantly greater proportion of those in cohort 1 (79.7% vs 66.7%) was successfully followed-up (P<.02), but those retained and lost to follow-up did not differ significantly in sociodemographic characteristics or medical or psychiatric comorbidity at inception. Table 1 presents the sociodemographic characteristics of patients who were followed-up. Cohort 1 hypochondriacal patients were of significantly lower social position than cohort 1 comparison patients. Cohort 2 hypochondriacal patients were significantly younger, of lower social position, and significantly more likely to be black, than cohort 2 comparison patients. For cohorts 1 and 2, the intrascale reliability coefficients (Cronbach α) were 0.91 to 0.95 for the Whiteley Index, 0.91 to 0.95 for the SSI, and 0.75 to 0.81 for the Somatosensory Amplification Scale. Given the similarity of the cohorts on the critical clinical variables, and the use of sociodemographic variables as covariates, combining cohorts 1 and 2 for subsequent analyses was considered justified. Age, educational level, and medical comorbidity at inception were then used as covariates in the multivariate analyses.

The picture at 4- to 5-year follow-up is presented in Table 2. The hypochondriacal group improved considerably, but 54 (63.5%) of the 85 hypochondriacal patients still met DSM-III-R diagnostic criteria for hypochondriasis. The chance-corrected concordance of the diagnosis of hypochondriasis (κ) at inception and follow-up was 0.62 (P<.0001). Hypochondriacal symptoms and somatization declined significantly during the interval, while role impairment improved significantly. Axis I psychiatric comorbidity declined slightly but not significantly, and Axis II caseness decreased significantly. Hypochondriacal patients developed an average of 0.3 new major medical problems and 0.88 moderate medical problems over the follow-up interval. Four hypochondriacal patients (3.3%) died. In the comparison group (Table 2), hypochondriacal symptoms remained stable except for a significant increase in bodily preoccupation; somatization increased very slightly; Axis I disorder and Axis II caseness remained stable; and role impairment showed a consistent trend toward improvement. Three comparison patients met diagnostic criteria for hypochondriasis at follow-up. Comparison patients developed an average of 0.4 major medical problems and 0.8 moderate problems over the 4- to 5-year interval, and 10 patients (7.5%) died. At follow-up (Table 2), the hypochondriacal group had significantly more hypochondriacal symptoms, somatization, amplification, psychiatric disorder, and impairment of role function than the comparison group. Their medical status did not differ significantly. The higher mortality rate in the comparison group was not statistically significant.

The changes in the 2 groups over time were then compared using MANOVA with repeated measures, and using age, educational level, and psychiatric and medical comorbidity as covariates. When the Whiteley Index and SSI were used as the dependent variables, a statistically significant, multivariate group×time interaction was found (Wilks λ=0.73, F_2,101=19.0, P <.001), as were significant univariate effects for the Whiteley Index (F_2,102=34.8, P<.001) and the SSI (F_2,102=22.6, P <.001). There were no significant differences between hypochondriacal and comparison patients in the incidence of medical illnesses (between-subjects MANOVA). There was a statistically significant multivariate group×time interaction for psychiatric morbidity (combining the index of Axis I disorder and Personality Diagnostic Questionnaire–Revised as the best indicator of overall psychiatric disturbance) (Wilks λ=0.94, F_2,103=3.1, P<.05), and significant univariate effect for the PDQ only (F_1,104=5.9, P<.015). A similar MANOVA with repeated measures of the Functional Status Questionnaire revealed a significant overall effect for group×time (Wilks λ=0.92, F_3,96=2.7, P<.05), and significant univariate effects for intermediate (F_1,98=4.5, P<.04) and social activities (F_1,98=6.6, P<.01), with the patients with hypochondriasis showing significantly greater improvements. A significant interaction effect (Wilks λ=0.93, F_2,99=3.5, P<.03) was also found for the number of days in bed and disability days, and significant univariate effects for both bed days (F_1,100=4.2, P<.04), and disability days (F_1,100=4.8, P <.03). Although both groups showed a decline over time, the decrease was significantly greater for hypochondriacal than comparison patients on both measures.

Thirty-one patients (36.5%) no longer met diagnostic criteria for hypochondriasis at follow-up (termed "remitting" patients or "remitters"). At inception (Table 3), remitters had significantly less disease conviction and less somatization than nonremitters. The same proportion of hypochondriacal patients with (32%) and without (37%) major medical comorbidity at inception remitted over the ensuing 5 years. The proportion of hypochondriacal patients with and without Axis I comorbidity at inception who remitted was not significantly different (32% [n=8] vs 44% [n=22], respectively). Remitters and nonremitters are compared at follow-up in Table 4. Remitters had significantly less somatization and amplification but did not differ significantly in role impairment, with the exception of significantly fewer disability days. The incidence of major medical illness during the follow-up interval was significantly greater in the remitter group.

Repeated measures analysis of covariance was used to compare these changes over time, while controlling for age and educational level. Remitters had significantly larger decreases in somatization than nonremitters (F_1,81=3.9, P<.05). There were no statistically significant differences between remitters and nonremitters on intermediate or social activities. Between-subjects MANOVA (controlling for age and educational level) revealed an overall statistically significant difference for major, moderate, and minor medical comorbidity (Wilks λ=0.84, F_3,71=4.5, P<.005). Univariate results indicate that remitters had significantly more major medical illnesses over the follow-up period (F_3,71=8.1, P<.006).

A number of caveats are in order. First, the patients who were followed-up were younger, less medically ill, and less geographically mobile than those lost to follow-up and this introduces the possibility of sampling bias. Second, cohorts 1 and 2 did not undergo the identical research battery. As a result, we do not have data from the entire sample for some of the measures. Third, we do not have information about intercurrent psychiatric treatment, either for hypochondriasis or for other comorbid conditions, nor can we assess the degree to which standard primary medical care may have been itself an effective treatment for hypochondriasis in some cases. Fourth, there is a tautological element in the study design: since the DSM diagnosis of hypochondriasis requires a minimum duration of 6 months, all eligible patients by definition had a relatively persistent condition.¹⁷ Fifth, our measure of medical morbidity has limitations. It could result in an overestimate of morbidity at follow-up since initial medical diagnoses that subsequently became inactive were unlikely to be deleted from the patient's problem list. There is also an issue of interrater reliability since different individuals conducted the ratings at inception to follow-up. The validity of the measure may also be questioned since aggregate medical morbidity increased over the follow-up period while disability declined. We suspect, however, that this is explained by the fact that patients were initially enrolled at the time of a clinic visit, when they were likely to be in an illness episode. This was less likely at follow-up.

Finally, there are doubts about the stability and reliability of the structured diagnostic interview method in general, and of the DIS in particular. Considerable temporal instability and unreliability has been reported when such interviews are readministered years apart, and when the results of lay interviewers and psychiatrists are compared.^38-41 Subject variance, information variance, and interviewer variance all pose a threat, and the first in particular may be quite substantial over long periods.⁴² This problem is confounded by the transition from DSM-III to DSM-III-R nosology that occurred during the course of this study. The diagnostic criteria for hypochondriasis were identical at inception and follow-up, however, and moderate concordance between inception and follow-up diagnosis was obtained.

Over a 4-to 5-year period, a sample of hypochondriacal patients experienced a significant reduction in symptoms and improvement in role functioning. By contrast, nonhypochondriacal patients exhibited little change in hypochondriacal symptoms and showed a modest, generally nonsignificant, improvement in role functioning. At follow-up, the hypochondriacal sample remains substantially more hypochondriacal, has more somatization and amplification, and is more functionally impaired and more psychiatrically disordered than the comparison group. The medical status of the 2 groups is similar. When these changes are examined in multivariate fashion, simultaneously taking into account baseline values and potential confounders, the hypochondriacal patients are seen to decline more in hypochondriacal symptoms, somatization, and disability than the comparison group. The changes in psychiatric and medical comorbidity, however, do not differ in the 2 groups.

Axis I psychiatric comorbidity is common in hypochondriasis.¹⁴ The prevalence of psychiatric disorder remained relatively stable over time in this sample, but methodological problems make these findings uncertain. Because of changes in the DSM and DIS interview over the course of the study, we constructed an index of psychiatric morbidity that was more nearly comparable at baseline and follow-up. It is still possible, however, that these problems obscured changes in psychiatric comorbidity, leading to a type II error.

In an earlier, cross-sectional study, we found that hypochondriacal and comparison patients did not differ significantly in aggregate medical morbidity.³⁷ We now find that the morbidity of both groups increased over the ensuing 4 to 5 years, but they do not differ significantly in this respect and the mortality of the hypochondriacal sample was one half that of the comparison group.

We have suggested elsewhere that bodily amplification is a relatively stable and enduring trait, which may predispose to hypochondriasis and somatization.^13,17,34,43 In the current study, amplification scores did not change over 4 to 5 years, even as hypochondriacal symptoms and somatization decreased. This confirms the traitlike character of amplification and suggests the possibility that even those hypochondriacal patients who remit (discussed below) may be vulnerable to relapse since they remain amplifiers of bodily sensation.

Some decline in the severity of hypochondriasis could be attributable to statistical regression toward the mean, but one would not expect this phenomenon to be large. One might also expect some decline in functional status over 4 to 5 years because of aging, yet no such decline was evident in the comparison group. This may reflect the fact (noted above) that patients were more likely to be in an illness episode at the inception interview than when the follow-up interviews were conducted.

Patients who subsequently remitted were slightly less hypochondriacal and had significantly less somatization at inception than those who continued to meet diagnostic criteria at follow-up. Remitters also exhibited a consistent but nonsignificant trend toward less disability at inception. Statistical power for these analyses was low (approximately 0.65 assuming a medium effect size), however, and may obscure meaningful predictors of course. At follow-up, remitters and nonremitters differed significantly only in that remitters had acquired more major medical diagnoses during the follow-up interval. The MANOVA with repeated measures confirmed that remission was more closely associated with a decline in somatization than in hypochondriacal attitudes, and that remitters had a significantly higher incidence of major medical diagnoses during the follow-up period.

The findings are open to contrasting interpretations. On the one hand, there is substantial stability in that hypochondriacal patients remain 1.5 to 2 times as hypochondriacal as the comparison group and two thirds continue to meet diagnostic criteria after a long interval. On the other hand, hypochondriacal symptoms and disability declined significantly on average, and a sizable minority no longer met diagnostic criteria. The unstructured interviews included in the research battery are instructive here. These interviews focused on the patient's clinical course during the follow-up period. Although impressionistic and qualitative, they suggested that clinical improvement was often precarious and tenuous because it appeared to result from external circumstances and environmental factors more than from a cognitive shift in the patients' views about their health. Many reported benefitting from fortuitous changes in their life circumstances that often seemed likely to change further, with the consequent potential of exacerbating their hypochondriasis. Other remitters had substituted elaborate, complicated, self-treatment regimens for their former, extensive pursuit of medical care. They allayed their disease concerns with an armamentarium of self-treatments (eg, diets, vitamin pills, exercise regimens, yoga, acupuncture, massage) which they regulated, adjusted, and combined depending on how they were feeling currently. Although these regimens provided a comforting sense of self-control over health, they were extensive, time consuming, and preoccupying, and the patient's clinical equilibrium appeared tenuous and fragile.

The qualitative interviews support the quantitative findings in suggesting that serious medical illness sometimes ameliorated hypochondriacal symptoms because it served to legitimize the patients' complaints, sanction their assumption of the sick role, affirm their experience of illness, and lessen the skepticism with which they had previously been regarded. Several patients described a sense of vindication and validation after becoming medically ill, and they noted an improvement in their relationships with their physicians. (As one noted, "Now that I know Dr X is paying attention to me, I can believe him if he says nothing serious is wrong.") Alternatively, those patients who ultimately became sick could have been misdiagnosed initially, and the somatic symptoms and disease fears originally called hypochondriacal could actually have been attributable to the medical disease that subsequently emerged. Based on our case-by-case reviews, however, we believe this occurred extremely infrequently, if at all.

Hypochondriasis thus appears to be a disorder with a substantial burden of long-term morbidity, functional impairment, and personal distress.

Accepted for publication March 24, 1998.

This investigation was supported by research grant MH-40487 from the National Institute of Mental Health, Bethesda, Md.

Reprints: Arthur J. Barsky, MD, Division of Psychiatry, Brigham 7 Women's Hospital, 75 Francis St, Boston, MA 02115 (e-mail: ajbarsky@bics.bwh.harvard.edu).