Charting patients' course: a comparison of statistics used to summarize patient course in longitudinal and repeated measures studies

Value and limitations of the BAR-score for donor allocation in liver transplantation

PURPOSE

The MELD-score was shown to be able to predict 90-day mortality in most patients with end-stage liver disease prior to liver transplantation and is used as a widely accepted measure for transplantation urgency. Prognostic ability of the BAR-score to predict 90-day post-transplant mortality by detection of unfavourable pretransplant combinations of donor and recipient factors may help to better balance urgency versus utility.

METHODS

Two German cohorts (Hannover, n=453; Kiel, n=234) were retrospectively analyzed using ROC-curve analysis, goodness-of-model-fit tests, summary measures and risk-adjusted multivariate binary regression. Included were all consecutive liver transplants performed in adult recipients (minimum age 18 years). Excluded were all combined transplants and living-related organ donor transplants.

RESULTS

Risk-adjusted multivariate regression revealed that the BAR-score is an independent risk factor for 90-day mortality after transplantation in both cohorts from Hannover and Kiel combined (p<0.001, OR=1.017, 95% CI:1.031-1.113). The area under the ROC-curve (AUROC) for the prediction of 90-day mortality using the BAR-score was 0.662 (95% CI 0.624-0.699, power>95%). Measures for association between observed 90-day mortality and the predicted probabilities in the combined cohort were concordant in 63.5% with low summary measures (Somers' D test 0.32, Goodman-Kruskal Gamma test 0.34 and Kendall's Tau a test 0.07).

CONCLUSIONS

The BAR-score performed below accepted thresholds for potentially useful clinical prognostic models. Prognostic models with better predictive ability with AUROCs>0.700, concordance>70% and larger summary measures are required for the prediction of 90-day post-transplant mortality to enable donor organ allocation with reliable weighing of urgency versus utility.

Intraindividual cognitive decline using a brief computerized cognitive screening test

BACKGROUND

Progressive intraindividual decline in memory and cognition is characteristic of dementia and may be useful in detecting very early Alzheimer's disease pathology.

METHODS

This study evaluated the slopes of cognitive performance over a 12-month period in 263 healthy, community-dwelling, adult volunteers aged ≥50 years. Participants completed a brief computerized battery of cognitive tests (CogState) at baseline and during 3-, 6-, 9-, and 12-month follow-up assessments. Linear mixed models were used to estimate age-adjusted mean slopes and 95% confidence intervals of change for each of the cognitive measures.

RESULTS

By defining age-adjusted mean slopes, and 95% confidence intervals for a measure of episodic memory, individuals with greater than expected decline (equal to or lower than the fifth percentile level of decline) were identified. From these, four individuals completed a full medical, neurologic, and neuropsychological evaluation, with none of them fulfilling criteria for mild cognitive impairment, but three (75%) having positive amyloid-positron emission tomographic scans.

CONCLUSIONS

Intraindividual decline in cognitive performance can be detected in otherwise healthy, community-dwelling, older persons, and this may deserve further study as a potential indicator of early Alzheimer's disease pathology.

On summary measure analysis of linear trend repeated measures data: performance comparison with two competing methods

Background

The summary measure approach (SMA) is sometimes the only applicable tool for the analysis of repeated measurements in medical research, especially when the number of measurements is relatively large. This study aimed to describe techniques based on summary measures for the analysis of linear trend repeated measures data and then to compare performances of SMA, linear mixed model (LMM), and unstructured multivariate approach (UMA).

Methods

Practical guidelines based on the least squares regression slope and mean of response over time for each subject were provided to test time, group, and interaction effects. Through Monte Carlo simulation studies, the efficacy of SMA vs. LMM and traditional UMA, under different types of covariance structures, was illustrated. All the methods were also employed to analyze two real data examples.

Results

Based on the simulation and example results, it was found that the SMA completely dominated the traditional UMA and performed convincingly close to the best-fitting LMM in testing all the effects. However, the LMM was not often robust and led to non-sensible results when the covariance structure for errors was misspecified. The results emphasized discarding the UMA which often yielded extremely conservative inferences as to such data.

Conclusions

It was shown that summary measure is a simple, safe and powerful approach in which the loss of efficiency compared to the best-fitting LMM was generally negligible. The SMA is recommended as the first choice to reliably analyze the linear trend data with a moderate to large number of measurements and/or small to moderate sample sizes.

Effect of antidepressants on the course of disability following stroke

OBJECTIVE

Stroke often produces marked physical and cognitive impairments leading to functional dependence, caregiver burden, and poor quality of life. We examined the course of disability during a 1-year follow-up period after stroke among patients who were administered antidepressants for 3 months compared to patients given placebo for 3 months.

METHODS

A total of 83 patients entered a double-blind randomized study of the efficacy of antidepressants to treat depressive disorders and reduce disability after stroke. Patients were assigned to either fluoxetine (N = 32), nortriptyline (N = 22) or placebo (N = 29). Psychiatric assessment included administration of the Present State Examination modified to identify DSM-IV symptoms of depression. The severity of depression was measured using the 17-item Hamilton Depression Rating Scale. The modified Rankin Scale was used to evaluate the disability of patients at initial evaluation and at quarterly follow-up visits for 1 year. Impairment in activities of daily living was assessed by Functional Independence Measure at the same time.

RESULTS

During the 1-year follow-up period, and after adjusting for critical confounders including age, intensity of rehabilitation therapy, baseline stroke severity, and baseline Hamilton Depression Rating Scale, patients who received fluoxetine or nortriptyline had significantly greater improvement in modified Rankin Scale scores compared to patients who received placebo (t [156] = -3.17, p = 0.002).

CONCLUSIONS

Patients treated with antidepressants had better recovery from disability by 1-year post stroke (i.e., 9 months after antidepressants were stopped) than patients who did not receive antidepressant therapy. This effect was independent of depression suggesting that antidepressants may facilitate the neural mechanisms of recovery in patients with stroke.

Using cognitive decline in novel trial designs for primary prevention and early disease-modifying therapy trials of Alzheimer's disease

BACKGROUND

Ideally putative disease-modifying therapies for Alzheimer's disease (AD) should be tested in patients who have minimal morbidity. Current barriers to such trials in early disease include the lack of disease-specific early biomarkers, insensitivity of quantitative cognitive outcome measures, and expensive trial designs requiring large sample sizes and long duration. This paper describes principles and progress towards a novel trial design that overcomes these problems, utilizing wide-scale cognitive performance screening to define pre-trial cognitive decline trajectories which can serve as trial outcome measures to assess AD disease-modifying efficacy.

METHODS

Theoretical principles important for the detection of intra-individual cognitive decline and a practical example are described.

RESULTS

Serial evaluations of community-based volunteers demonstrate how a screening tool method to detect subtle cognitive decline can predict in vivo amyloid pathology as a trigger for etiological evaluation. Trajectories of decline appear consistent over at least two years, suggesting they could be used as a trial inclusion criterion and ameliorable outcome measure together with other AD biomarkers. Informative trial durations could be 6-12 months, or extend to incorporate staggered random withdrawal or start designs, with as few as 20 individuals per treatment arm.

CONCLUSIONS

This trial methodology offers significant advantages over current AD trial designs, including treatment at earlier stages of disease, shorter trial duration, obviation of informed consent difficulties, smaller sample sizes, reduced cost and--given adequate screening programs--sufficient subjects for multiple simultaneous trials. Importantly, it allows the rapid evaluation of putative treatments that may only be efficacious in pre-dementia states.

Decline in verbal memory in non-demented older adults

Decline in memory function was detected in 30% of healthy community-dwelling elderly over 6 years using a task assessing delayed word list recall. Individuals with memory decline over time also demonstrated relative deficits on additional tasks of memory and learning, a task of working memory and executive function, and on a verbal (category) fluency task at their most recent assessment. These relative deficits in the performance of individuals with memory decline cannot be explained by age-related changes, education, intelligence, mood, health-related factors, or the individuals' APOE epsilon 4 status. Decline in memory performance did not result in greater complaints of cognitive difficulties when compared with normal elderly, nor did it limit overall participation in life activities. Although the significance of memory decline in the current study was not determined quantitatively, memory decline is consistent with the early deterioration characteristic of mild cognitive impairment and preclinical Alzheimer's disease and confirms the need to monitor individuals with objective memory decline, even when these individuals fall within normal limits for a given neuropsychological task.

Concurrent tracking of alcohol use and bipolar disorder symptoms

OBJECTIVES

Alcohol use disorders (AUDs) are common co-occurring conditions in patients with bipolar disorder (BD), but it is unclear whether or not AUD and BD symptoms are temporally correlated. The primary aim of this analysis was to examine concurrent symptom tracking and how the relative onsets of AUD and BD influence the concurrent tracking of symptoms.

METHODS

Participants met DSM-IV criteria for bipolar I disorder, manic or mixed, with no prior hospitalizations and minimal treatment. Patients were rated for alcohol use and bipolar symptom severity on a weekly basis for up to 7 years. For analysis purposes, patients were placed into groups with no AUD (BD Only; n = 21), onset of AUD either concurrent with or after the onset of bipolar symptoms (BD First; n = 32), and onset of AUD at least 1 year before the onset of bipolar symptoms (AUD First; n = 18).

RESULTS

None of the patient groups demonstrate consistent positive or negative temporal correlations between alcohol use and affective symptoms. However, there were significant between-group differences on the relationship of symptom tracking and age of BD onset. For the AUD First group, the correlation between age of BD onset and symptom tracking was positive 0.41. However, for the BD First and BD Only groups the correlations were negative (-0.32 and -0.41, respectively). Moreover, for patients whose BD onset was < or =18 years old, the correlation between age of onset and tracking was -0.47.

CONCLUSIONS

These findings suggest that although there is no direct temporal correlation of AUD and BD symptoms in subgroups of BD patients, age at illness onset contributes to the complex relationship between BD and AUD. For younger patients there may be a greater likelihood that alcohol use and bipolar symptoms increase or decrease in unison.

More powerful two-sample tests for differences in repeated measures of adverse effects in psychiatric trials when only some patients may be at risk

Common adverse effect measures in psychiatric trials are typically analysed with repeated measures ANOVA, despite having distributions which violate key assumptions of that method; moreover, some adverse effects may be concentrated in vulnerable subgroups of participants. For testing treatment differences in adverse effects, we propose use of Kendall's taub as a summary measure of within-participant trends in adverse events, in conjunction with a weighted modification of a rank test proposed by Conover and Salsburg. Data on extrapyramidal side effects from a controlled clinical trial conducted in persons with treatment resistant schizophrenia was used to compare the proposed analysis to repeated measures ANOVA using mixed models and alternate tests for treatment differences in taub trend scores.

Acute treatment outcomes in patients with bipolar I disorder and co-morbid borderline personality disorder receiving medication and psychotherapy

OBJECTIVE

Patients suffering from both bipolar I disorder and borderline personality disorder (BPD) pose unique treatment challenges. The purpose of this matched case-control study was to compare acute treatment outcomes of a sample of patients who met standardized diagnostic criteria for both bipolar I disorder and BPD (n = 12) to those who met criteria for bipolar I disorder only (n = 58).

METHOD

Subjects meeting criteria for an acute affective episode were treated with a combination of algorithm-driven pharmacotherapy and weekly psychotherapy until stabilization (defined as four consecutive weeks with a calculated average of the 17-item version of the Hamilton Rating Scale for Depression and Bech-Rafaelsen Mania scale totaling < or = 7).

RESULTS

Only three of 12 (25%) bipolar-BPD patients achieved stabilization, compared with 43 of 58 (74%) bipolar-only patients. Two of the three bipolar-BPD patients who did stabilize took over 95 weeks to do so, compared with a median time-to-stabilization of 35 weeks in the bipolar-only group. The bipolar-BPD group received significantly more atypical mood-stabilizing medications per year than the bipolar-only group (Z = 4.3, p < 0.0001). Dropout rates in the comorbid group were high.

CONCLUSIONS

This quasi-experimental study suggests that treatment course may be longer in patients suffering from both bipolar I disorder and BPD. Some patients improved substantially with pharmacotherapy and psychotherapy, suggesting that this approach is worthy of further investigation.

Catching up on health outcomes: the Texas Medication Algorithm Project

OBJECTIVE

To develop a statistic measuring the impact of algorithm-driven disease management programs on outcomes for patients with chronic mental illness that allowed for treatment-as-usual controls to "catch up" to early gains of treated patients.

DATA SOURCES/STUDY SETTING

Statistical power was estimated from simulated samples representing effect sizes that grew, remained constant, or declined following an initial improvement. Estimates were based on the Texas Medication Algorithm Project on adult patients (age > or = 18) with bipolar disorder (n = 267) who received care between 1998 and 2000 at 1 of 11 clinics across Texas.

STUDY DESIGN

Study patients were assessed at baseline and three-month follow-up for a minimum of one year. Program tracks were assigned by clinic.

DATA COLLECTION/EXTRACTION METHODS

Hierarchical linear modeling was modified to account for declining-effects. Outcomes were based on 30-item Inventory for Depression Symptomatology-Clinician Version.

PRINCIPAL FINDINGS

Declining-effect analyses had significantly greater power detecting program differences than traditional growth models in constant and declining-effects cases. Bipolar patients with severe depressive symptoms in an algorithm-driven, disease management program reported fewer symptoms after three months, with treatment-as-usual controls "catching up" within one year.

CONCLUSIONS

In addition to psychometric properties, data collection design, and power, investigators should consider how outcomes unfold over time when selecting an appropriate statistic to evaluate service interventions. Declining-effect analyses may be applicable to a wide range of treatment and intervention trials.

A staging approach to measuring patient-centred subjective outcomes

INTRODUCTION

In assessing clinical change, measurement is often based on psychometric scales. However, change is best revealed within the constellation of problems salient to the patient, rather than in alterations in the abstract constructs, psychometrically measured. These patients' problems often serially unfold in qualitative stages, even before the full-blown disorder emerges. These qualitative stages constitute the natural history extending from early to late, fluctuating from mild to severe, and progressing from full-blown disorder to recovery.

METHOD

We reviewed the literature on clinimetrics and patient-centred subjective measures, and related these findings to the use of the discretized-analogue scaling method.

RESULTS

There is increasing recognition of clinimetric approaches that structure the pre-clinical and clinical material into a scale that reflects the symptoms, consequences and complications in a manner understandable to the patient, and enabling the quantification of severity or change. This monograph provides criteria and methods for developing these building blocks that enable the assessment of severity, stage or change. We show examples of their use in quantitative clinical outcome measurement.

CONCLUSION

We encourage further studies in the ideology and procedures for measuring clinical change in terms of personally subjective experiences.

Determining the extent of cognitive change after coronary surgery: a review of statistical procedures

Currently, cognitive decline after coronary surgery is said to be significant if the individual's postoperative test score is at least 1 standard deviation (SD) worse than their preoperative score. This "1-SD" technique fails to account for factors that may confound interpretation of serially acquired cognitive test scores, including regression to the mean, measurement error caused by poor test-retest reliability, and practice effects. We review the many alternative and potentially superior statistical techniques that have been described in the neuropsychologic and psychiatric literature for differentiating "true" changes in cognitive test score from changes caused by these factors.

The effect of remission of poststroke depression on activities of daily living in a double-blind randomized treatment study

Poststroke depression has been associated with impaired recovery of activities of daily living (ADL) during the first 2 years after stroke. This study examined the effect of remission of poststroke depression on recovery in ADL in a double-blind randomized treatment study. Based on a semistructured psychiatric exam and DSM-IV diagnostic criteria, a consecutive series of 23 patients who met criteria for major depression (N = 16) or minor depression (N = 7) were selected and randomly assigned to either active treatment (nortriptyline) or placebo. Functional physical (i.e., ADL) impairment was assessed using the Johns Hopkins Functioning Inventory (JHFI). Patients whose depressive disorder remitted at follow-up had significantly greater recovery in ADL functions compared with patients whose depression did not remit. There were no differences in demographic variables, lesion characteristics, and neurological symptoms between the two groups, which would explain the significantly greater improvement among the remitted patients. Because both major and minor depression patients who remitted showed greater improvement in ADL than nonremitted patients some of whom were treated with active and some with placebo medication, nonpharmacotherapeutic mechanisms related to recovery from depression appear to mediate this enhanced recovery.

Adding subjects or adding measurements: Which increases the precision of longitudinal research?

When designing repeated measurement studies, researchers must strike a balance between increasing the number of subjects and increasing the number of measurement times for each subject. The question often becomes "Do I gain more statistical precision by adding subjects or by adding additional follow-up measurements?" This study presents a method for evaluating the relative benefit of adding subjects versus adding measurement times. We used the standard error of estimate (SE) for mean change as the criterion of precision. An existing dataset on post-stroke patients containing six follow-up assessments of six standard rating scales was used. SE values for two common change indices were found and compared for all possible two, three, four, five, and six repeated measurements. Sample sizes required to achieve the same benefit as adding an additional measurement are presented. These data suggest that collecting five or six repeated measurements may be sufficient for accurately assessing change and that attempts to further precision should be accomplished by increasing the sample size.