Motivation: Selecting a small number of relevant genes for accurate classification of samples is essential for the development of diagnostic tests. We present the Bayesian model averaging (BMA) method for gene selection and classification of microarray data. Typical gene selection and classification procedures ignore model uncertainty and use a single set of relevant genes (model) to predict the class. BMA accounts for the uncertainty about the best set to choose by averaging over multiple models (sets of potentially overlapping relevant genes).

Results: We have shown that BMA selects smaller numbers of relevant genes (compared with other methods) and achieves a high prediction accuracy on three microarray datasets. Our BMA algorithm is applicable to microarray datasets with any number of classes, and outputs posterior probabilities for the selected genes and models. Our selected models typically consist of only a few genes. The combination of high accuracy, small numbers of genes and posterior probabilities for the predictions should make BMA a powerful tool for developing diagnostics from expression data.

Availability: The source codes and datasets used are available from our Supplementary website.

Contact: kayee@u.washington.edu

Supplementary information: http://www.expression.washington.edu/publications/kayee/bma

/pdf/bayesian-model-averaging-development-of-an-improved-multi-3kpqykhpx9.pdf

Bayesian model averaging: development of an improved multi-class, gene selection and classification tool for microarray data

Breast cancer is a heterogeneous disease, though little is known about some of its rarer forms, including certain histologic types. Using Surveillance, Epidemiology, and End Results Program data on 135 157 invasive breast cancer cases diagnosed from 1992 to 2001, relationships between nine histologic types of breast cancer and various tumour characteristics were assessed. Among women aged 50–89 years at diagnosis, lobular and ductal/lobular carcinoma cases were more likely to be diagnosed with stage III/IV, ⩾5.0 cm, and node-positive tumours compared to ductal carcinoma cases. Mucinous, comedo, tubular, and medullary carcinomas were less likely to present at an advanced stage. Lobular, ductal/lobular, mucinous, tubular, and papillary carcinomas were less likely, and comedo, medullary, and inflammatory carcinomas were more likely to be oestrogen receptor (ER) negative/progesterone receptor (PR) negative and high grade (notably, 68.2% of medullary carcinomas were ER−/PR− vs 19.3% of ductal carcinomas). In general, similar differences were observed among women diagnosed at age 30–49 years. Inflammatory carcinomas are associated with more aggressive tumour phenotypes, and mucinous, tubular, and papillary tumours are associated with less aggressive phenotypes. The histologic types of breast cancer studied here differ greatly in their clinical presentations, and the differences in their hormone receptor profiles and grades point to their likely different aetiologies.

/pdf/clinical-characteristics-of-different-histologic-types-of-whceauuci8.pdf

Clinical characteristics of different histologic types of breast cancer.

Objective To investigate the incidence, transition probabilities, and risk factors for obsessive-compulsive disorder (OCD) and subclinical OCD in adolescents. Method A two-stage epidemiological study originally designed to investigate depression was conducted between 1987 and 1989 in the southeastern United States. For the screening, a self-report depressive symptom questionnaire was administered to a community sample of 3,283 adolescents. In the diagnostic stage, the Schedule for Affective Disorders and Schizophrenia for School-Age Children was administered to 488 mother–child pairs. Baseline screening and diagnostic data from the first year the subject completed an interview and follow-up diagnostic data from subsequent years were used. Results The 1-year incidence rates of OCD and subclinical OCD were found to be 0.7% and 8.4%, respectively. Transition probabilities demonstrated a pattern of moving from more severe to less severe categories. Of those with baseline OCD, 17% had the diagnosis of OCD at follow-up; 62% moved to the referent group. Of those with baseline subclinical OCD, 1.5% had OCD at follow-up and 75% moved to the referent group. Black race (odds ratio [OR] = 23.38), age (OR = 4.02), desirable life events (OR = 0.78), undesirable life events (OR = 1.21), and socioeconomic status (OR not estimable) were significant predictors of incident OCD. Age (OR = 2.30), desirable life events (OR = 0.92), and undesirable life events (OR = 1.13) were significantly associated with incident subclinical OCD. Conclusion An initial diagnosis of subclinical OCD was not significantly predictive of a diagnosis of OCD at 1-year follow-up. The overall morbidity remained higher at follow-up in the baseline OCD group than in the baseline subclinical OCD group. The baseline subclinical OCD group was more dysfunctional at follow-up than was the baseline referent group. Further research concerning differences in symptomatology and impairment between OCD and subclinical OCD is warranted.

/pdf/frequency-of-obsessive-compulsive-disorder-in-a-community-15rvh8f446.pdf

Frequency of obsessive-compulsive disorder in a community sample of young adolescents.

Social scientists are commonly interested in relating a latent trait (e.g., criminal tendency) to measurable individual covariates (e.g., poor parenting) to understand what defines or perhaps causes the latent trait. In this article we develop an efficient and convenient method for answering such questions. The basic model presumes that two types of variables have been measured: Response variables (possibly longitudinal) that partially determine the latent class membership, and covariates or risk factors that we wish to relate to these latent class variables. The model assumes that these observable variables are conditionally independent, given the latent class variable. We use a mixture model for the joint distribution of the observables. We apply this model to a longitudinal dataset assembled as part of the Cambridge Study of Delinquent Development to test a fundamental theory of criminal development. This theory holds that crime is committed by two distinct groups within the population: Adoles...

Modeling Uncertainty in Latent Class Membership: A Case Study in Criminology

ContextWomen using combined estrogen and progestin hormone replacement therapy
(CHRT) have an increased risk of breast cancer; however, data on use for long
durations and on risk associated with patterns of use are lacking.ObjectiveTo evaluate relationships between durations and patterns of CHRT use
and risk of breast cancer by histological type and hormone receptor status.DesignPopulation-based case-control study.SettingThree counties in western Washington State.ParticipantsNine hundred seventy-five women 65-79 years of age diagnosed with invasive
breast cancer from April 1, 1997, through May 31, 1999 (histology: 196 lobular
cases, 656 ductal cases, 114 cases with other histological type, and 9 cases
with an unspecified histological type; estrogen receptor (ER)/progesterone
receptor (PR) status: 646 ER+/PR+ cases, 147 ER+/PR− cases, and 101
ER−/PR− cases [6 ER−/PR+ cases and 75 cases with unknown
ER/PR status were not included in the analyses herein]) and 1007 population
controls.Main Outcome MeasuresRisks of invasive lobular, ductal, ER+/PR+, ER+/PR−, and ER−/PR−
breast carcinomas.ResultsWomen using unopposed estrogen replacement therapy (ERT) (exclusive
ERT use), even for 25 years or longer, had no appreciable increase in risk
of breast cancer, although the associated odds ratios were not inconsistent
with a possible small effect. Ever users of CHRT (includes CHRT users who
also had used ERT) had a 1.7-fold (95% confidence interval [CI], 1.3-2.2)
increased risk of breast cancer, including a 2.7-fold (95% CI, 1.7-4.3) increased
risk of invasive lobular carcinoma, a 1.5-fold (95% CI, 1.1-2.0) increased
risk of invasive ductal carcinoma, and a 2.0-fold (95% CI, 1.5-2.7) increased
risk of ER+/PR+ breast cancers. The increase in risk was greatest in those
using CHRT for longer durations (users for 5-14.9 years and ≥15 years had
1.5-fold [95% CI, 1.0-2.3] and 1.6-fold [95% CI, 1.0-2.6] increases in risk
of invasive ductal carcinoma, respectively, and 3.7-fold [95% CI, 2.0-6.6]
and 2.6-fold [95% CI, 1.3-5.3] increases in risk of invasive lobular carcinoma,
respectively. Associations of similar magnitudes were seen among users of
both sequential and continuous CHRT. Risks of ER+/PR− and ER−/PR−
tumors were not increased by use of any form of hormone replacement therapy;
however, small numbers of these tumors limited power to detect possible associations.ConclusionThese data suggest that use of CHRT is associated with an increased
risk of breast cancer, particularly invasive lobular tumors, whether the progestin
component was taken in a sequential or in a continuous manner.

https://jamanetwork.com/journals/jama/articlepdf/196803/joc30501.pdf

Relationship between long durations and different regimens of hormone therapy and risk of breast cancer

SUMMARY The use of individualized logistic regression, in which a series of separate simple logistic regression analyses are performed as a replacement for polychotomous logistic regression, is studied. The asymptotic relative efficiencies of the individual parameter estimates are observed to be generally high, as are the efficiencies of predicted probability estimates and, to a somewhat lesser extent, joint tests of parameters from different regressions.

Colin B. Becg

Papers

Calculation of polychotomous logistic regression parameters using individualized regressions