The effect of missing data and imputation on the detection of bias in cognitive testing using differential item functioning methods.

BACKGROUND: Item response theory (IRT) methods for addressing differential item functioning (DIF) can detect group differences in responses to individual items (e.g., bias). IRT and DIF-detection methods have been used increasingly often to identify bias in cognitive test performance by characteristics (DIF grouping variables) such as hearing impairment, race, and educational attainment. Previous analyses have not considered the effect of missing data on inferences, although levels of missing cognitive data can be substantial in epidemiologic studies. METHODS: We used data from Visit 6 (2016-2017) of the Atherosclerosis Risk in Communities Neurocognitive Study (N = 3,580) to explicate the effect of artificially imposed missing data patterns and imputation on DIF detection. RESULTS: When missing data was imposed among individuals in a specific DIF group but was unrelated to cognitive test performance, there was no systematic error. However, when missing data was related to cognitive test performance and DIF group membership, there was systematic error in DIF detection. Given this missing data pattern, the median DIF detection error associated with 10%, 30%, and 50% missingness was -0.03, -0.08, and -0.14 standard deviation (SD) units without imputation, but this decreased to -0.02, -0.04, and -0.08 SD units with multiple imputation. CONCLUSIONS: Incorrect inferences in DIF testing have downstream consequences for the use of cognitive tests in research. It is therefore crucial to consider the effect and reasons behind missing data when evaluating bias in cognitive testing.

Assessing Bias in Cognitive Testing for Older Adults with Sensory Impairment: An Analysis of Differential Item Functioning in the Baltimore Longitudinal Study on Aging (BLSA) and the Atherosclerosis Risk in Communities Neurocognitive Study (ARIC-NCS).

OBJECTIVES: Vision and hearing impairments affect 55% of people aged 60+ years and are associated with lower cognitive test performance; however, tests rely on vision, hearing, or both. We hypothesized that scores on tests that depend on vision or hearing are different among those with vision or hearing impairments, respectively, controlling for underlying cognition. METHODS: Leveraging cross-sectional data from the Baltimore Longitudinal Study of Aging (BLSA) and the Atherosclerosis Risk in Communities Neurocognitive Study (ARIC-NCS), we used item response theory to test for differential item functioning (DIF) by vision impairment (better eye presenting visual acuity worse than 20/40) and hearing impairment (better ear .5-4 kHz pure-tone average > 25 decibels). RESULTS: We identified DIF by vision impairment for tests whose administrations do not rely on vision [e.g., Delayed Word Recall both in ARIC-NCS: .50 logit difference between impaired and unimpaired (p = .04) and in BLSA: .62 logits (p = .02)] and DIF by hearing impairment for tests whose administrations do not rely on hearing [Digit Symbol Substitution test in BLSA: 1.25 logits (p = .001) and Incidental Learning test in ARIC-NCS: .35 logits (p = .001)]. However, no individuals had differences between unadjusted and DIF-adjusted measures of greater than the standard error of measurement. CONCLUSIONS: DIF by sensory impairment in cognitive tests was independent of administration characteristics, which could indicate that elevated cognitive load among persons with sensory impairment plays a larger role in test performance than previously acknowledged. While these results were unexpected, neither of these samples are nationally representative and each has unique selection factors; thus, replication is critical.

'Alzheimer's Progression Score': Development of a Biomarker Summary Outcome for AD Prevention Trials.

BACKGROUND: Alzheimer's disease (AD) prevention research requires methods for measurement of disease progression not yet revealed by symptoms. Preferably, such measurement should encompass multiple disease markers. OBJECTIVES: Evaluate an item response theory (IRT) model-based latent variable Alzheimer Progression Score (APS) that uses multi-modal disease markers to estimate pre-clinical disease progression. DESIGN: Estimate APS scores in the BIOCARD observational study, and in the parallel PREVENT-AD Cohort and its sister INTREPAD placebo-controlled prevention trial. Use BIOCARD data to evaluate whether baseline and early APS trajectory predict later progression to MCI/dementia. Similarly, use longitudinal PREVENT-AD data to assess test measurement invariance over time. Further, assess portability of the PREVENT-AD IRT model to baseline INTREPAD data, and explore model changes when CSF markers are added or withdrawn. SETTING: BIOCARD was established in 1995 and participants were followed up to 20 years in Baltimore, USA. The PREVENT-AD and INTREPAD trial cohorts were established between 2011-2015 in Montreal, Canada, using nearly identical entry criteria to enroll high-risk cognitively normal persons aged 60+ then followed for several years. PARTICIPANTS: 349 cognitively normal, primarily middle-aged participants in BIOCARD, 125 high-risk participants aged 60+ in PREVENT-AD, and 217 similar subjects in INTREPAD. 106 INTREPAD participants donated up to four serial CSF samples. MEASUREMENTS: Global cognitive assessment and multiple structural, functional, and diffusion MRI metrics, sensori-neural tests, and CSF concentrations of tau, Aß42 and their ratio. RESULTS: Both baseline values and early slope of APS scores in BIOCARD predicted later progression to MCI or AD. Presence of CSF variables strongly improved such prediction. A similarly derived APS in PREVENT-AD showed measurement invariance over time and portability to the parallel INTREPAD sample. CONCLUSIONS: An IRT-based APS can summarize multimodal information to provide a longitudinal measure of pre-clinical AD progression, and holds promise as an outcome for AD prevention trials.

Intraprostatic ethanol chemoablation via transurethral and transperineal injection.

OBJECTIVES: To further assess the safety and feasibility of prostatic chemoablation with ethanol and to address previous concerns associated with transperineal injection using a canine model. MATERIALS AND METHODS: The study included 25 dogs; normal saline or 98% dehydrated ethanol were injected into the prostate using both routes, at volumes of 25-50% of the total prostate volume. The prostate and adjacent structures were examined grossly and histopathologically after the dogs were killed humanely at 4 h, 7 days and 12 weeks after injection. RESULTS: Transperineal injection resulted in tissue necrosis in all prostates and significant extraprostatic necrosis in two of three animals treated. With transurethral injection, the control groups showed minimal change, whereas the group injected with ethanol resulted in lesions with variable necrosis and location. CONCLUSIONS: Intraprostatic chemoablation is possible with ethanol injection both transperineally and transurethrally. Transperineal ethanol injections were associated with more extraprostatic necrosis. Transurethral injections resulted in larger amounts of necrosis in the prostatic parenchyma with minimal extraprostatic effects. However, the extent of prostatic necrosis/ablation was inconsistent and further research is warranted.

Bayesian Interval Estimation of Multiple Correlations with Missing Data: A Gibbs Sampling Approach.

A Bayesian method for obtaining an interval estimate of the population squared multiple correlation from an incomplete multivariate normal data set is described. The method is applicable to data sets where values are missing on any combination of the dependent and independent variables. Further, the missing data need not be missing in a completely random fashion. The estimates are constructed using a Markov Chain Monte Carlo procedure known as Gibbs Sampling. The important issues of the convergence properties of the Gibbs sampler, the effect of the choice of a reference prior, and the empirical coverage probabilities of the estimates are considered in detail. Investigations using simulated data suggest that the proposed method can yield accurate interval estimates of the population squared multiple correlation.

Evaluation of biofeedback seat insert for improving active sitting posture in children with cerebral palsy. A clinical report.

Biofeedback devices have been used successfully to improve head control and symmetrical standing in children with cerebral palsy. This clinical report describes a biofeedback seat insert developed to improve erect sitting posture in children with cerebral palsy who have inadequate trunk control. The seat insert is easily placed against the back of any seating device. A momentary-contact pressure switch on the seat insert is activated when the child exerts pressure on it by extending his trunk. The pressure switch then activates a videocassette recorder or can be adapted to activate a television or radio. Five children with spastic cerebral palsy participated in this evaluation of the biofeedback seat insert. The results of this evaluation show that the children used the biofeedback seat insert effectively to actively improve their sitting posture by voluntarily extending their trunk against the pressure switch. The biofeedback seat insert offers physical therapists a valuable therapeutic training tool to encourage carry-over of improved sitting posture away from the clinical setting for children with cerebral palsy.

A COMPARISON OF THE PREDICTIVE ACCURACY OF A POOLING AND A SUBGROUPING PREDICTION STRATEGY.

The subgrouping strategy has been employed in a number of studies (Fredriksen & Gilbert, 1960; Fredriksen & Melville, 1964; Ghiselli, 1956; Klein, Rock & Evans, 1968; Rock, Evans & Klein, 1969; Malnig, 1969). In each of these studies it was shown that after partitioning a sample into g subgroups and performing a separate regression analysis within each group, one could identify one or more subgroups having a larger multiple correlation than the total sample regression. An important question is whether this type of result will be maintained under cross validation.

A MONTE CARLO STUDY OF THE ACCURACY OF A HIERARCHICAL GROUPING PROCEDURE.

The accuracy of the Ward-Hook method of hierarchical grouping was studied using artificial data sampled from a mixture distribution. The expected number of sample misclassifications as well as the variance of these mimlassifications were taken as the criteria of subgrouping accuracy. It was found that the expected sample misclassificatiun rate was a positively biased estimator of the population probability of misclassifioartion. Further, by in- c d n g sample size, the variance of the sample misclassifications decreased. Accurate subgroupings were found in geneml when the population misclassification rake was very low (.02). When this probability was increased to .20, the misclassification variance increased sharply, unless sample size was large.