Constantine Frangakis

The Prevention and Treatment of Missing Data in Clinical Trials

Missing data in clinical trials can have a major effect on the validity of the inferences that can be drawn from the trial. This article reviews methods for preventing missing data and, failing that, dealing with data that are missing.

Multiple imputation by chained equations: what is it and how does it work?

Multivariate imputation by chained equations (MICE) has emerged as a principled method of dealing with missing data. Despite properties that make MICE particularly useful for large imputation procedures and advances in software development that now make it accessible to many researchers, many psychiatric researchers have not been trained in these methods and few practical resources exist to guide researchers in the implementation of this technique. This paper provides an introduction to the MICE method with a focus on practical aspects and challenges in using this method. A brief review of software programs available to implement MICE and then analyze multiply imputed data is also provided. Copyright

Principal stratification approach to broken randomized experiments: A case study of school choice vouchers in New York City

The precarious state of the educational system in the inner cities of the United States, as well as its potential causes and solutions, have been popular topics of debate in recent years. Part of the difficulty in resolving this debate is the lack of solid empirical evidence regarding the true impact of educational initiatives. The efficacy of so-called “school choice” programs has been a particularly contentious issue. A current multimillion dollar program, the School Choice Scholarship Foundation Program in New York, randomized the distribution of vouchers in an attempt to shed some light on this issue. This is an important time for school choice, because on June 27, 2002 the U.S. Supreme Court upheld the constitutionality of a voucher program in Cleveland that provides scholarships both to secular and religious private schools. Although this study benefits immensely from a randomized design, it suffers from complications common to such research with human subjects: noncompliance with assigned “treatments” and missing data. Recent work has revealed threats to valid estimates of experimental effects that exist in the presence of noncompliance and missing data, even when the goal is to estimate simple intention-to-treat effects. Our goal was to create a better solution when faced with both noncompliance and missing data. This article presents a model that accommodates these complications that is based on the general framework of “principal stratification” and thus relies on more plausible assumptions than standard methodology. Our analyses revealed positive effects on math scores for children who applied to the program from certain types of schools—those with average test scores below the citywide median. Among these children, the effects are stronger for children who applied in the first grade and for African-American children.

Effect of Citalopram on Agitation in Alzheimer Disease: The CitAD Randomized Clinical Trial

IMPORTANCE Agitation is common, persistent, and associated with adverse consequences for patients with Alzheimer disease. Pharmacological treatment options, including antipsychotics are not satisfactory. OBJECTIVE The primary objective was to evaluate the efficacy of citalopram for agitation in patients with Alzheimer disease. Key secondary objectives examined effects of citalopram on function, caregiver distress, safety, cognitive safety, and tolerability. DESIGN, SETTING, AND PARTICIPANTS The Citalopram for Agitation in Alzheimer Disease Study (CitAD) was a randomized, placebo-controlled, double-blind, parallel group trial that enrolled 186 patients with probable Alzheimer disease and clinically significant agitation from 8 academic centers in the United States and Canada from August 2009 to January 2013. INTERVENTIONS Participants (n = 186) were randomized to receive a psychosocial intervention plus either citalopram (n = 94) or placebo (n = 92) for 9 weeks. Dosage began at 10 mg per day with planned titration to 30 mg per day over 3 weeks based on response and tolerability. MAIN OUTCOMES AND MEASURES Primary outcome measures were based on scores from the 18-point Neurobehavioral Rating Scale agitation subscale (NBRS-A) and the modified Alzheimer Disease Cooperative Study-Clinical Global Impression of Change (mADCS-CGIC). Other outcomes were based on scores from the Cohen-Mansfield Agitation Inventory (CMAI) and the Neuropsychiatric Inventory (NPI), ability to complete activities of daily living (ADLs), caregiver distress, cognitive safety (based on scores from the 30-point Mini Mental State Examination [MMSE]), and adverse events. RESULTS Participants who received citalopram showed significant improvement compared with those who received placebo on both primary outcome measures. The NBRS-A estimated treatment difference at week 9 (citalopram minus placebo) was -0.93 (95% CI, -1.80 to -0.06), P = .04. Results from the mADCS-CGIC showed 40% of citalopram participants having moderate or marked improvement from baseline compared with 26% of placebo recipients, with estimated treatment effect (odds ratio [OR] of being at or better than a given CGIC category) of 2.13 (95% CI, 1.23-3.69), P = .01. Participants who received citalopram showed significant improvement on the CMAI, total NPI, and caregiver distress scores but not on the NPI agitation subscale, ADLs, or in less use of rescue lorazepam. Worsening of cognition (-1.05 points; 95% CI, -1.97 to -0.13; P = .03) and QT interval prolongation (18.1 ms; 95% CI, 6.1-30.1; P = .01) were seen in the citalopram group. CONCLUSIONS AND RELEVANCE Among patients with probable Alzheimer disease and agitation who were receiving psychosocial intervention, the addition of citalopram compared with placebo significantly reduced agitation and caregiver distress; however, cognitive and cardiac adverse effects of citalopram may limit its practical application at the dosage of 30 mg per day. TRIAL REGISTRATION clinicaltrials.gov Identifier: NCT00898807.

A Mouse Model of Blast Injury to Brain: Initial Pathological, Neuropathological, and Behavioral Characterization

The increased use of explosives in recent wars has increased the number of veterans with blast injuries. Of particular interest is blast injury to the brain, and a key question is whether the primary overpressure wave of the blast is injurious or whether brain injury from blast is mostly due to secondary and tertiary effects. Using a shock tube generating shock waves comparable to open-field blast waves, we explored the effects of blast on parenchymatous organs of mice with emphasis on the brain. The main injuries in nonbrain organs were hemorrhages in the lung interstitium and alveolar spaces and hemorrhagic infarcts in liver, spleen, and kidney. Neuropathological and behavioral outcomes of blast were studied at mild blast intensity, that is, 68 ± 8 kPag (9.9 ±1.2 psig) static pressure, 103 kPag (14.9 psig) total pressure and 183 ± 14 kPag (26.5 ± 2.1 psig) membrane rupturepressure. Under these conditions, weobserved multifocal axonal injury, primarily in the cerebellum/brainstem, the corticospinal system, and the optic tract. We also found prolonged behavioral and motor abnormalities, including deficits in social recognition and spatial memory and in motor coordination. Shielding of the torso ameliorated axonal injury and behavioral deficits. These findings indicate that long CNS axon tracts are particularly vulnerable to the effects of blast, even at mild intensities that match the exposure of most veterans in recent wars. Prevention of some of these neurological effects by torso shielding may generate new ideas as to how to protect military and civilian populations in blast scenarios.

The consistency statement in causal inference: A definition or an assumption?

Three assumptions sufficient to identify the average causal effect are consistency, positivity, and exchangeability (ie, “no unmeasured confounders and no informative censoring,” or “ignorability of the treatment assignment and measurement of the outcome”). The exchangeability assumptions are well known territory for epidemiologists and biostatisticians. Briefly, to be satisfied, these 2 exchangeability assumptions that require exposed and unexposed subjects, and censored and uncensored subjects have equal distributions of potential outcomes, respectively. Indeed, the so-called fundamental problem of causal inference is directly linked to the first exchangeability assumption. In contrast, the consistency and positivity assumptions are less well known. The positivity assumption states that there is a nonzero (ie, positive) probability of receiving every level of exposure for every combination of values of exposure and confounders that occur among individuals in the population. It remains unclear why the consistency and positivity assumptions are less well known. Optimistically, perhaps these assumptions are less important with respect to an impact on estimation of the average causal effect. Pessimistically, these assumptions are less well known because there is little alarming evidence of a departure from either of these assumptions in observational studies without explicitly looking for the departure. Here we will focus on the preliminary issue of clarifying the consistency assumption. The consistency assumption is often stated such that an individual’s potential outcome under her observed exposure history is precisely her observed outcome. Methods for causal inference require that the exposure is defined unambiguously. Specifically, one needs to be able to explain how a certain level of exposure could be hypothetically assigned to a person exposed to a different level. This requirement is known as consistency. Consistency is guaranteed by design in experiments, because application of the exposure to any individual is under the control of the investigator. Consistency is plausible in observational studies of medical treatments, because one can imagine how to manipulate hypothetically an individual’s treatment status. However, consistency is problematic in observational studies with exposures for which manipulation is difficult to conceive. Consistency is especially difficult when the exposure is a biologic feature, such as body weight, insulin resistance, or CD4 cell count. For example, there are many competing ways to assign (hypothetically) a body mass index of 25 kg/m to an individual, and each of them may have a different causal effect on the outcome. To state consistency formally, let us first define individual j’s potential outcome Yj(x) under exposure x as the outcome that would have been observed if individual j had received exposure x. The variable Yj(x) is known as a potential outcome because it

A role of sunshine in the triggering of suicide

Several reports indicate that suicide follows a seasonal pattern with a dominant peak during the month of maximum daylight. The purpose of this study was to evaluate the hypothesis that sunshine exposure may trigger suicidal behavior. We found a remarkably consistent pattern of seasonality with peak incidence around June in the northern hemisphere and December in the southern hemisphere. Moreover, there was a positive association between the seasonal amplitude of suicide (measured by relative risk) and total sunshine in the corresponding country. These findings indicate that sunshine may have a triggering effect on suicide, and suggests further research in the field of sunshine-regulated hormones, particularly melatonin.

The relationship of excess body weight and health- related quality of life : evidence from a population study in Taiwan

Objective:Excess body weight is related to significant morbidity and mortality. However, less is known about the relationship of body weight to health-related quality of life (HRQOL), especially for Asian populations. We examined the relationship of excess weight and HRQOL in a general population sample from Taiwan.Research methods and procedures:This cross-sectional study used a national representative sample (n=14 221) from the 2001 Taiwan National Health Interview Survey. Body weight was categorized using body mass index (BMI in kg/m2) as normal (18.5–24.9), overweight (25–29.9), and obese (⩾30). HRQOL was measured using the Taiwan version of the SF-36. We compared the body weight–HRQOL relationships by age, gender, and status of chronic condition, respectively. We especially used the Generalized Estimating Equations (GEE) to examine the relationships of BMI and HRQOL by taking into account the correlations of HRQOL within households. Four models were developed to adjust sequentially for sets of covariates: Model 1 with no adjustment; Model 2 adjusting for sociodemographic variables; Model 3 adding chronic conditions; Model 4 further adding smoking status.Results:Unadjusted physical HRQOL was best for normal weight, worse for overweight, and worst for obese individuals. For unadjusted mental HRQOL, overweight subjects had at least as good mental domain scores of HRQOL as those with normal weight or obesity, depending on the subscales. As age increased, excess weight was associated with worse physical, but not mental HRQOL. Compared to men, women with excess weight showed a greater deficit in physical HRQOL. Multivariable analyses suggested that obesity was associated with worse physical HRQOL compared to overweight, which, in turn, was worse or comparable to normal weight. Specifically, in the model adjusting for demographic variables, the deficit in physical functioning and physical component scores for the obese vs normal weight were statistical significant (P<0.05) and clinically important difference (effect size ⩾0.3). Both obesity and overweight were associated with higher mental component scores than normal weight, but the effect size was <0.3.Conclusion:In Taiwan, excess weight was related to worse physical, but not mental HRQOL. The lack of impact of increased body weight on mental health status presents a potential challenge to preventing the increases in obesity. More research is needed to elucidate the mechanisms by which excess weight affects specific domains of HRQOL, and to develop effective prevention strategies.

Multiple Imputation With Large Data Sets: A Case Study of the Children's Mental Health Initiative

Multiple imputation is an effective method for dealing with missing data, and it is becoming increasingly common in many fields. However, the method is still relatively rarely used in epidemiology, perhaps in part because relatively few studies have looked at practical questions about how to implement multiple imputation in large data sets used for diverse purposes. This paper addresses this gap by focusing on the practicalities and diagnostics for multiple imputation in large data sets. It primarily discusses the method of multiple imputation by chained equations, which iterates through the data, imputing one variable at a time conditional on the others. Illustrative data were derived from 9,186 youths participating in the national evaluation of the Community Mental Health Services for Children and Their Families Program, a US federally funded program designed to develop and enhance community-based systems of care to meet the needs of children with serious emotional disturbances and their families. Multiple imputation was used to ensure that data analysis samples reflect the full population of youth participating in this program. This case study provides an illustration to assist researchers in implementing multiple imputation in their own data.

Do the SF-36 and WHOQOL-BREF measure the same constructs? Evidence from the Taiwan population*

Background: The SF-36 and WHOQOL-BREF are available for international use, but it is not clear if they measure the same constructs. We compared the psychometric properties and factor structures of these two instruments. Methods: Data were collected from a national representative sample (n=11,440) in the 2001 Taiwan National Health Interview Survey, which included Taiwan versions of the SF-36 and WHOQOL-BREF. We used Cronbach’s alpha coefficient to estimate scale reliability. We conducted exploratory factor analysis to determine factor structure of the scales, and applied multitrait analysis to evaluate convergent and discriminant validity. We used standardized effect size to compare known-groups validity for health-related variables (including chronic conditions and health care utilization) and self-reported overall quality of life. Structural equation modeling was used to analyze relationships among the two SF-36 component scales (PCS and MCS) and the four WHOQOL subscales (physical, psychological, social relations, and environmental). Results: Cronbach’s alpha coefficients were acceptable (⩾0.7) for all subscales of both instruments. The factor analysis yielded two unique factors: one for the 8 SF-36 subscales and a second for the 4 WHOQOL subscales. Pearson correlations were weak (<0.3) among subscales of both instruments. Correlations for subscales hypothesized to measure similar constructs differed little from those measuring heterogeneous subscales. Effect sizes suggested greater discrimination by the SF-36 for health status and services utilization known groups, but greater discrimination by the WHOQOL for QOL-defined groups. Structural equation modeling suggested that the SF-36 PCS and MCS were weakly associated with WHOQOL. Conclusions: In this Taiwan population sample, the SF-36 and WHOQOL-BREF appear to measure different constructs: the SF-36 measures health-related QOL, while the WHOQOL-BREF measures global QOL. Clinicians and researchers should carefully define their research questions related to patient-reported outcomes before selecting which instrument to use.