“Predicting In-Hospital Mortality of Patients with AIDS-Related Pneumocystis carinii Pneumonia: An Example of Hierarchically Optimal Classification Tree Analysis” (Yarnold et al., Statistics in Medicine, 1997, 16, 1451-1463): Corrected Illustration of CTA Model

Paul R. Yarnold

Optimal Data Analysis, LLC

The lead note in the Errata section of Optimal Data Analysis presents the corrected illustration of the first CTA model published in the field of medicine, two decades ago.

View journal article

Type A Behavior and Savoring Among College Undergraduates: Enjoy Achievements Now—Not Later

Fred B. Bryant & Paul R. Yarnold

Optimal Data Analysis, LLC

Recent research tested the a priori hypothesis that Type A Behavior (TAB) undermines enjoyment of leisure time, and that this effect is mediated by savoring responses which hamper enjoyment. Findings suggested that the hypothesized A-B differences in savoring reflect differences in perfectionism rather than in time urgency. The present study uses the same sample to compare 117 extreme Type A and 131 extreme B undergraduates on ten dimensions of savoring assessed for a performance-related stimulus. Findings revealed Type As focus on how proud they are and impressed others are, but are only moderately to weakly involved in actively storing positive memories for later recall, or in reminiscing about prior positive events.

View journal article

Hierarchically Optimal Classification Tree Analysis of Adverse Drug Reactions Secondary to Warfarin Therapy

Robert C. Soltysik & Paul R. Yarnold

Optimal Data Analysis, LLC

Identification and detection of adverse drug reactions (ADRs) is critical to patient safety improvement, and warfarin is a medication known to be associated with a high rate of ADRs. Hierarchically optimal CTA was used to predict ADRs attributable to medications taken in addition to warfarin, for a sample of 2,289 hospital in-patients. A model involving five medications was identified, with relatively high ADR rates identified for patients, already on warfarin, who also received zolpidem tartrate, tamsulosin HCL, famotidine, nitroglycerin, and rofecoxib. The CTA model achieved moderate classification accuracy (ESS=38.0), correctly classifying 1,323 of 2,246 patients (58.9%) without an ADR, and 34 of 43 patients (79.1%) experiencing an ADR.

View journal article

Emergency Severity Index (Version 3) Score Predicts Hospital Admission

Paul R. Yarnold & Robert C. Soltysik

Optimal Data Analysis, LLC

Hospital admission status, Emergency Severity Index (ESI) Version 3 triage score, and binary indicators of whether lab work or radiological examinations were completed in the Emergency Department (ED), were available for 160,471 patients seen over a three-year period in the ED of a leading community teaching hospital in Toronto. Hierarchical CTA conducted for these data yielded ESS=63.8, a relatively strong effect. The model correctly classified 80.2% of the admitted patients, and 83.6% of the patients who weren’t admitted. The model was correct 60.1% of the time that it predicted that a patient would be admitted, and correct 93.2% of the time that it predicted that a patient wouldn’t be admitted.

View journal article

UniODA vs. Logistic Regression: Serum Cholesterol and Coronary Heart Disease and Mortality Among Middle Aged Diabetic Men

Paul R. Yarnold

Optimal Data Analysis, LLC

The effect of serum cholesterol level on coronary heart disease and mortality was assessed for middle aged diabetic men in a prospective population study.1 Logistic regression was used to test the linear trend over quintiles, yielding estimated p<0.02. However the validity of the estimated Type I error rate is called into question because the minimum expectation for chi-square was violated. UniODA was applied to these data and identified a model yielding moderate classification accuracy (ESS=29.9, exact p<0.0006), which was stable in jackknife validity analysis.

View journal article

UniODA vs. Kappa: Evaluating the Long-Term (27-Year) Test-Retest Reliability of the Type A Behavior Pattern

Paul R. Yarnold

Optimal Data Analysis, LLC

This 27-year follow-up investigated long-term stability of Type A behavior (TAB) for 1,180 surviving participants in the Western Collaborative Group Study. The kappa statistic was used to assess reliability among and between self- and Structured Interview-based TAB assessments. Results indicated fair temporal reliability for self-assessments (kappa=0.39), moderate temporal reliability for Interview-assessments (0.24), and low parallel-forms reliability for self and Interview assessments at intake (0.16) or follow-up (0.11). In contrast, for these data UniODA found relatively strong temporal reliability for self-assessments (ESS=50.2), and weak effects for all other estimates (ESS<23.3).

View journal article

UniODA vs. Weighted Kappa: Evaluating Concordance of Clinician and Patient Ratings of the Patient’s Physical and Mental Health Functioning

Paul R. Yarnold

Optimal Data Analysis, LLC

This study investigated the concordance between clinician and patient assessments of patient’s physical and mental functioning, made using 4-category ordinal scales, for a consecutive sample of 166 outpatients with rheumatoid arthritis. Weighted kappa isn’t a normed statistic, but the respective weighted kappa statistic obtained for the assessments, 0.39 and 0.30, are believed to reflect fair concordance. Data were also analyzed via confirmatory UniODA models which hypothesized that physician and patient ratings agreed. Findings indicated relatively strong concordance for the physical health ratings (ESS=55.5), and moderate concordance for mental health ratings (ESS=43.3).

View journal article

UniODA vs. Chi-Square: Discriminating Inhibited and Uninhibited Infant Profiles

Paul R. Yarnold

Optimal Data Analysis, LLC

Kagan and Snidman investigated processes mediating early reactivity to stimulation in a longitudinal study of 94 four-month-old infants who displayed a combination of either high motor activity and frequent crying, or low motor activity and infrequent crying. Fearful behavior assessed at 9 and 14 months of age was examined in relation to these two infant typologies. Eyeball analysis, which was confirmed statistically using chi-square analysis, revealed that 40% of low motor activity infants displayed “low fear” (which was arbitrarily defined as one or fewer fears) at both 9 and 14 months, versus 0% of high motor activity infants. When UniODA was applied to these data it identified statistically reliable effects at 9- and 14-months: the strongest effect occurred at 14 months. Applying CTA to these data revealed that a multiattribute model wasn’t feasible.

View journal article

UniODA vs. Student’s t-Test: Comparing Two Migraine Treatments

Paul R. Yarnold

Optimal Data Analysis, LLC

This study evaluates the number of migraine attacks experienced in a clinical trial of two alternative treatments, for a sample of 67 patients. Several conventional statistical methods were used to compare the number of attacks between treatments, but all of these methods were compromised by violations of their underlying assumptions. Only an exact test motivated by an eyeball spline was revealing. UniODA was used to compare treatments, and identified the latter effect.

View journal article