Clustering and variable selection in the presence of mixed variable types and missing data.

We consider the problem of model-based clustering in the presence of many correlated, mixed continuous, and discrete variables, some of which may have missing values. Discrete variables are treated with a latent continuous variable approach, and the Dirichlet process is used to construct a mixture model with an unknown number of components. Variable selection is also performed to identify the variables that are most influential for determining cluster membership. The work is motivated by the need to cluster patients thought to potentially have autism spectrum disorder on the basis of many cognitive and/or behavioral test scores. There are a modest number of patients (486) in the data set along with many (55) test score variables (many of which are discrete valued and/or missing). The goal of the work is to (1) cluster these patients into similar groups to help identify those with similar clinical presentation and (2) identify a sparse subset of tests that inform the clusters in order to eliminate unnecessary testing. The proposed approach compares very favorably with other methods via simulation of problems of this type. The results of the autism spectrum disorder analysis suggested 3 clusters to be most likely, while only 4 test scores had high (>0.5) posterior probability of being informative. This will result in much more efficient and informative testing. The need to cluster observations on the basis of many correlated, continuous/discrete variables with missing values is a common problem in the health sciences as well as in many other disciplines.

Evaluation of the ability of neurobiological, neurodevelopmental and socio-economic variables to predict cognitive outcome in premature infants.

BACKGROUND: When developed in the 1990s, the Neurobiologic Risk Score (NBRS) and Neurodevelopmental Risk Exam (NRE) correlated well with developmental outcomes in premature infants. Given recent advances in neonatology, we assessed their present ability to predict cognitive outcome, alone and combined with socio-economic factors. METHODS: One hundred and twenty-nine neonates <32 weeks gestational age were assessed at 6, 12 and/or 24 months corrected age with the Cognitive Adaptive Test/Clinical Linguistic and Auditory Milestone Scale (CAT/CLAMS). Indices of socio-economic status included maternal education and marital status. RESULTS: At 24 months corrected age (n= 67), the NBRS (r=-0.5), maternal education (r= 0.46) and marital status (r= 0.37) correlated with the CAT/CLAMS. These correlations increased when NBRS and maternal education were combined (r= 0.63) and when specific NBRS components (intraventricular haemorrhage, periventricular leukomalacia, seizures) and maternal education were combined (r= 0.66). CONCLUSIONS: In the contemporary neonatal intensive care unit, measures used to predict cognitive outcome should incorporate both neurobiological risk factors and socio-economic variables.

Longitudinal direct medical costs associated with constipation in women.

BACKGROUND: Although direct medical costs for constipation-related medical visits are thought to be high, to date, there have been no studies examining longitudinal resource utilisation in adults with constipation. AIM: To estimate the incremental direct medical costs associated with constipation in women. METHODS: This is a nested case-control study. The study population consisted of all mothers of 5718 children in the population-based birth cohort born during 1976-1982 in a community. The cases presented to the medical facilities with constipation. The controls were randomly selected and matched to cases in a 2:1 ratio. Direct medical costs for constipated women and controls were collected for the years 1987-2002. RESULTS: We identified 168 women with a diagnosis of constipation. The total direct medical costs over the 15-year period for constipated subjects were more than double those of controls [$63 591 (95% CI: 49 786-81 396) vs. $24 529 (95% CI: 20 667-29 260)]. The overall out-patient costs for constipated women were $38 897 (95% CI: 31 381-48 253) compared to $15 110 (95% CI: 12 904-17 781) for controls. The median of annual out-patient visits for constipated women was 0.16 compared to 0.11 for controls. CONCLUSION: Women with constipation have significantly higher medical care utilisation and expenditures compared with women without constipation.

Incidence of reading disability in a population-based birth cohort, 1976-1982, Rochester, Minn.

OBJECTIVE: To report the incidence of reading disability among school-aged children. SUBJECTS AND METHODS: In this population-based, retrospective birth cohort study, subjects included all 5718 children born between 1976 and 1982 who remained in Rochester, Minn, after the age of 5 years. Based on records from all public and nonpublic schools, medical facilities, and private tutorial services and on results of all individually administered IQ and achievement tests, extensive medical, educational, and socioeconomic information were abstracted. Reading disability was established with use of research criteria based on 4 formulas (2 regression-based discrepancy, 1 non-regression-based discrepancy, and 1 low achievement). RESULTS: Cumulative incidence rates of reading disability varied from 5.3% to 11.8% depending on the formula used. Boys were 2 to 3 times more likely to be affected than girls, regardless of the identification methods applied. CONCLUSIONS: In this population-based birth cohort, reading disability was common among school-aged children and significantly more frequent among boys than girls, regardless of definition.

Boy/girl differences in risk for reading disability: potential clues?

The authors conducted a case-control study to determine whether risk factors for reading disability (RD) differentially affect boys and girls. The study population included all children born between 1976 and 1982 in Olmsted County, Minnesota (n = 5,701). A total of 303 RD cases were identified by using intelligence quotient and achievement test scores collected from school and medical records. After excluding those who met exclusion criteria (n = 869), controls consisted of all children not identified with RD (n = 4,529). The authors examined the association between RD and potential risk factors in boys and girls and confirmed their results in multivariable logistic regression models. Multivariable models indicated that girls of low birth weight were more than twice as likely to be identified as RD (odds ratio (OR) = 2.94, 95% confidence interval (CI): 1.09, 6.25). Girls whose mothers had 12 or fewer years of education were twice as likely to be identified as RD (OR = 2.14, 95% CI: 1.24, 3.72). However, girls whose fathers were aged 35 years or older at the time of birth were less likely to be identified as RD (OR = 0.24, 95% CI: 0.06, 0.92). Only 12 or fewer years of paternal education was associated with increased RD in boys (OR = 2.28, 95% CI: 1.59, 3.27). Boys and girls appear to be differentially susceptible to RD risk factors, suggesting that the biologic processes leading to RD may differ between boys and girls.

School attendance and school performance: a population-based study of children with asthma.

OBJECTIVE: To analyze school attendance and school achievement as outcomes of the care of children with asthma. METHODS: A previously identified Rochester, Minnesota, cohort of children with asthma and age- and sex-matched children without asthma were studied. School attendance, standardized achievement test scores, grade point average, grade promotion, and class rank of graduating students for children with asthma and control subjects were obtained from the Rochester Public School system. RESULTS: Children with asthma (n = 92) and age- and sex-matched non-asthmatic control subjects with 640 school-years of observation were studied. Children with asthma had 2.21 (95% CI, 1.41 to 3.01) more days absent than children without asthma. There was no significant difference in standardized achievement test scores (reading percentile difference 1.22% [95% CI, -3.68 to 6.12], mathematics percentile difference 2.36% [95% CI, -2.89 to 7.60], language percentile difference 2.96% [95% CI, -4.03 to 7.15]). There was no significant difference in grade point average, grade promotion, or class rank of graduating students. CONCLUSION: In this community, although children with asthma had 2 excess days of absenteeism, the school performance of children with asthma was similar to that of children without asthma.

Use and costs of medical care for children and adolescents with and without attention-deficit/hyperactivity disorder.

CONTEXT: A shortage of data exists on medical care use by persons with attention-deficit/hyperactivity disorder (ADHD). OBJECTIVE: To compare medical care use and costs among persons with and without ADHD. DESIGN AND SETTING: Population-based cohort study conducted in Rochester, Minn. SUBJECTS: All children born in 1976-1982 were followed up through 1995, using school and medical records to identify those with ADHD. The 4880 birth cohort members (mean age, 7. 3 years) still residing in Rochester in 1987 were followed up in medical facility-linked billing databases until death, emigration, or December 31, 1995. MAIN OUTCOME MEASURES: Clinical diagnoses, likelihood and frequency of inpatient and outpatient hospitalizations, emergency department (ED) visits, and total medical costs (including ambulatory care), compared among individuals with and without ADHD. RESULTS: Among the 4119 birth cohort members who remained in the area through 1995 (mean age, 15.3 years), 7.5% (n = 309) had met criteria for ADHD. Compared with persons without ADHD, those with ADHD were more likely to have diagnoses in multiple categories, including major injuries (59% vs 49%; P<.001) and asthma (22% vs 13%; P<.001). The proportion with any hospital inpatient, hospital outpatient, or ED admission was higher for persons with ADHD vs those without ADHD (26% vs 18% [P<. 001], 41% vs 33% [P =.006], and 81% vs 74% [P =.005], respectively). The 9-year median costs for persons with ADHD compared with those without ADHD were more than double ($4306 vs $1944; P<.001), even for the subset with no hospital or ED admissions (eg, median 1987 costs, $128 vs $65; P<.001). The differences between individuals with and without ADHD were similar for males and females and across all age groups. CONCLUSION: In our cohort, compared with persons without ADHD, those with ADHD exhibited substantially greater use of medical care in multiple care delivery settings.

Decline in prostate cancer mortality from 1980 to 1997, and an update on incidence trends in Olmsted County, Minnesota.

PURPOSE: We describe trends in prostate cancer mortality from 1980 to 1997, before and after the introduction of serum prostate specific antigen (PSA) testing to the community medical practice, and provide an update on trends in incidence since 1992. MATERIALS AND METHODS: All men with a diagnosis of prostate cancer who died between 1980 and 1997 were identified and parts 1 and 2 of the death certificates were reviewed for a diagnosis of prostate cancer. In addition, all men with biopsy proved prostate cancer diagnosed between 1983 and 1995 were identified. The complete medical records of incident cases of prostate cancer were reviewed for signs and symptoms at diagnosis and for the first treatment received. RESULTS: Age adjusted, community mortality rates from prostate cancer increased from 25.8/100,000 men in 1980 to 1984 to a peak of 34/100,000 in 1989 to 1992, and have since declined to 19.4/100,000 in 1993 to 1997 (22% decline in mortality, 95% confidence interval 49% decline to 17% increase). The overall age adjusted incidence rates which peaked at 209/100,000 person-years in 1992 as previously reported declined to 108/100,000 in 1993 and 132/100,000 in 1995. A similar pattern was observed for organ confined cancers. However, incidence rates for regional or distant disease were suggestive of a continuing downward trend from 1989 to 1992 compared to 1993 to 1995 (12% decline per year, p = 0.07). CONCLUSIONS: These data demonstrate that despite the increase in prostate cancer mortality rates in the mid to late 1980s, mortality rates in 1993 to 1997 are lower than in the years before serum PSA testing. While chance cannot be ruled out, the data suggest that increased screening for prostate cancer, particularly through PSA testing, may have led to declines in mortality from prostate cancer.

Potential influence of migration bias in birth cohort studies.

OBJECTIVE: To assess the potential bias, due to migration from the community, in a birth cohort study of learning disability in Olmsted County, Minnesota. MATERIAL AND METHODS: The 1976 through 1982 birth cohort consisted of 8,548 children born to mothers who were residents of Rochester, Minnesota, at the time of delivery. The current status of all children was determined for the 1995 to 1996 school year with the resources of Independent School District #535, the Rochester Epidemiology Project, and the Rochester Reading Center. Information from birth certificates was compared between migrants and nonmigrants with use of standard statistical tests. RESULTS: The cumulative probability of migration by age 5 years was 32.2% (95% confidence interval, 31.2 to 33.2%). When migrants were compared with nonmigrants, the distributions of Apgar scores, father's age, gestational age at birth, and number of children in the family were virtually identical. In contrast, migrants were more likely to be nonwhite (6% versus 2%), be born to a single parent (11% versus 7%), and have a congenital defect noted at birth (1.5% versus 0.7%). When considered simultaneously in a logistic regression model, the parents of migrants were more highly educated, migrant mothers were younger and had fewer prenatal visits, and migrant children were more likely to be black. CONCLUSION: This report demonstrates the extreme effort that must be made to assemble a birth cohort in our mobile society. Furthermore, the results suggest that only slight differences exist between migrants and nonmigrants; these are unlikely to compromise the future results of the ongoing learning disability study.

Screening digital rectal examination and prostate cancer mortality: a population-based case-control study.

OBJECTIVES: Although digital rectal examination (DRE) for the detection of prostate cancer has been recommended by many professional groups and has become part of the general physical examination, no randomized clinical trial has demonstrated the efficacy or effectiveness of this practice. We conducted a population-based case-control study to evaluate the association between DRE and prostate cancer mortality. METHODS: With the resources of the Rochester Epidemiology Project, all 173 men who died of prostate cancer in Olmsted County from 1976 to 1991, who were resident at the time of diagnosis, were identified. For each case, two control patients were drawn from the population, matched for residence at the time of diagnosis in the case, birth date, and duration of medical record in Olmsted County. Trained nurse abstractors reviewed the community medical records for up to 10 years before the date of diagnosis in the case for mention of DRE and specific findings associated with each mention. RESULTS: Case subjects were less likely than control subjects to have had any DRE in the 10 years before diagnosis (matched odds ratio [OR] = 0.51; 95% confidence interval [CI] = 0.31, 0.84). When limited to DREs without mention of signs or symptoms that might raise suspicion of prostate cancer, the association was even stronger (OR = 0.31; 95% CI = 0.19, 0.49). Adjustment for educational attainment, marital status, and comorbid conditions did little to alter the associations. CONCLUSIONS: These results demonstrate a strong inverse association between DRE and prostate cancer mortality. If further research concludes this association to be causal, screening DREs may have prevented as many as 50% to 70% of deaths due to prostate cancer that might have occurred in the absence of screening.

Secular changes in radical prostatectomy utilization rates in Olmsted County, Minnesota 1980 to 1995.

PURPOSE: We estimated the changes in utilization of radical prostatectomy for treatment of prostate cancer and describe the clinical characteristics of men undergoing radical prostatectomy in a population based setting. MATERIALS AND METHODS: The Rochester Epidemiology Project was used to identify all Olmsted County residents who underwent radical prostatectomy from 1980 to 1995. The community medical records of these men were reviewed to determine the clinical and pathological stage and grade at biopsy and following surgery. RESULTS: From 1980 to 1995, 311 radical prostatectomies were performed on Olmsted County men. From 1980 to 1987 prostatectomy rates ranged from 6.3 to 31.0/100,000 men but rates increased dramatically to 53.6/100,000 in 1988 and 106.2/100,000 in 1992. The rate after 1992 decreased to 53.0/100,000 and then increased slightly to 80.4/100,000. There was a shift to younger age in more recent times (mean patient age 65.4 years in 1980 to 1986 and 62.4 in 1993 to 1995, p = 0.02), a nonsignificant (p = 0.10) trend toward lower pathological stage in recent years (42% stage pT2 in 1980 to 1986 versus 55% in 1993 to 1995) and a significant decrease in the proportion of cases of disease up staged following surgery (53% in 1980 to 1986 versus 37% in 1993 to 1995, p = 0.03). There was no significant trend in pathological grade with time (63% Mayo grade I or II in 1980 to 1986 versus 52% in 1993 to 1995, p = 0.30). CONCLUSIONS: These findings demonstrate an increase in radical prostatectomy rates that coincided with increases in prostate cancer incidence. There was a decrease in population prostatectomy rates in 1993 which was followed by modest increases to levels lower than the peak in 1992. However, the clinical characteristics of patients during this period did not change dramatically, suggesting that in a population based setting the selection factors for patients undergoing surgical treatment may not have changed.

Predictive properties of serum-prostate-specific antigen testing in a community-based setting.

BACKGROUND: Most studies that have described the sensitivity and specificity of prostate-specific antigen (PSA) as a screening test have been conducted in urology practice settings or in media-based screening programs. The control patients from these settings may have a higher prevalence of urologic disorders that increase serum PSA levels than that of the general population in which screening efforts might take place, leading to biased estimates of sensitivity and specificity. OBJECTIVE: To determine the sensitivity and specificity of serum PSA levels for the early detection of prostate cancer in a population-based setting. PATIENTS AND METHODS: This population-based case-control study was conducted in Olmsted County, Minnesota, where the Rochester Epidemiology Project could identify all incident cases of prostate cancer through passive surveillance of medical care provided to local residents. Case patients were all 177 men (age range, 50-79 years) who were newly diagnosed as having prostate cancer from 1990 through 1992 and had a prediagnostic serum PSA determination (90% of all incident cases). Control patients were randomly selected from the Olmsted County population and had undergone a clinical examination to exclude prostate cancer. RESULTS: The median (25th and 75th percentiles) of serum PSA levels was 9.4 ng/mL (5.4 and 18.6 ng/mL, respectively) for case patients and 1.2 ng/mL (0.7 and 2.1 ng/mL, respectively) for control patients (P < .001). When sensitivity was plotted against 1-specificity, the area under the receiver operating characteristic curve was 0.94 (SE, 0.01). The predictive power declined somewhat with age, with areas under the curve of 0.96, 0.94, and 0.90 for men in their 50s, 60s, and 70s, respectively. When cases were restricted to the 155 men with clinically localized disease, the area under the curve was essentially unchanged (0.94; SE, 0.01) and still much greater than the estimates of 0.75 that were reported from urology practice- and media-based settings. CONCLUSIONS: In a community-based setting, serum PSA levels provide better discrimination between men with and without clinically localized prostate cancer than has been observed in studies that were conducted in urologic practices. These results suggest that previous decision analyses may have underestimated the predictive value of PSA for the detection of prostate cancer in a primary care or community-wide screening program.

Mental retardation in a birth cohort, 1976-1980, Rochester, Minnesota.

The cumulative incidence of mental retardation in a birth cohort of children born from 1976 through 1980 in Rochester, Minnesota (n = 5,919) was estimated. Rochester is the site of the Rochester Epidemiology Project, which captures virtually all medical care delivered locally. Passive follow-up through school and community medical records for criteria diagnostic of mild or severe mental retardation was undertaken. Thirty children were classified with mental retardation. The cumulative incidence of mental retardation by age 8 years was 9.1 per 1,000 (95% confidence interval = 6.2, 13.0) and was similar for boys and girls (8.3 vs. 10.0 per 1,000, respectively). The cumulative incidence for severe mental retardation in girls was more than twice that in boys; for mild retardation it was the opposite. The cumulative incidence among boys was 1.7 times greater than the cumulative incidence for girls. Results suggest that although the overall incidence of mental retardation by gender was similar, the severity may differ somewhat.

Incidence of prostate cancer diagnosis in the eras before and after serum prostate-specific antigen testing.

OBJECTIVE: To estimate the incidence of prostate cancer in Olmsted County, Minnesota, from 1983 through 1992 to describe the secular changes that have occurred since the introduction of serum prostate-specific antigen (PSA) testing to the community medical practice in 1987. DESIGN: Population-based, descriptive epidemiological study with ecological and individual level comparisons over time. STUDY SETTING: Olmsted County, Minnesota, where the Rochester Epidemiology Project provides passive surveillance of the population for health outcomes. SUBJECTS: All 511 biopsy-proven incident cases of adenocarcinoma of the prostate diagnosed from 1983 through 1992. The community inpatient and outpatient medical records of all incident cases were reviewed to evaluate the presenting characteristics of men at the time of diagnosis. RESULTS: The age-adjusted incidence of biopsy-proven prostate cancer increased from 64 per 100,000 person-years in 1983 to 216 per 100,000 person-years in 1992. The increase occurred primarily between 1987 and 1988 and was predominately for organ-confined tumors. The age-specific incidence increased dramatically in this same period among men aged 50 years and older. Among men aged 70 years and older, however, prostate carcinoma incidence rates declined after 1990 following the initial increase. This decline among older men contrasted with community-based estimates of PSA utilization rates, which demonstrated consistent increases since 1987 to nearly 50% of the older population in 1992. CONCLUSION: These results support the premise that the recent increase in prostate cancer is due in part to the increased utilization of serum PSA testing. Further, the increased incidence appears to be a transient phenomenon due to the depletion of previously undiagnosed cases from the previous pool. Finally, these data suggest that, in terms of stage at diagnosis, early detection efforts may be effective in identifying more early stage (smaller) cancers.

Incidence, clinical features, and prognosis in Bell's palsy, Rochester, Minnesota, 1968-1982.

The average annual incidence of Bell's palsy per 100,000 population in Rochester, Minnesota, for 1968 through 1982 was 25.0 for both sexes combined; crude rates for males and females were 22.8 and 26.9, respectively, based on 85 males and 121 females. The relationship between various clinical features, patient characteristics, and the type of recovery was analyzed. In 206 patients, 28 (14%) experienced incomplete recovery and 178 (86%) had complete recovery, based on evidence in the medical records. Using logistic regression, complete facial weakness, non-ear pain, and hypertension were identified as the most important risk factors for incomplete recovery. Patients were divided into two groups for comparison of treatment results; one group (n = 94) was without any of the three identified risk factors, and the other group (n = 112) consisted of patients who had one or more risk factors. Results suggested that among the latter group, those treated with steroids fared better than those in any of the other treatment groups.