David C Hoaglin

Conducting indirect-treatment-comparison and network-meta-analysis studies: Report of the ISPOR task force on indirect treatment comparisons good research practices: Part 2

Evidence-based health care decision making requires comparison of all relevant competing interventions. In the absence of randomized controlled trials involving a direct comparison of all treatments of interest, indirect treatment comparisons and network meta-analysis provide useful evidence for judiciously selecting the best treatment(s). Mixed treatment comparisons, a special case of network meta-analysis, combine direct evidence and indirect evidence for particular pairwise comparisons, thereby synthesizing a greater share of the available evidence than traditional meta-analysis. This report from the International Society for Pharmacoeconomics and Outcomes Research Indirect Treatment Comparisons Good Research Practices Task Force provides guidance on technical aspects of conducting network meta-analyses (our use of this term includes most methods that involve meta-analysis in the context of a network of evidence). We start with a discussion of strategies for developing networks of evidence. Next we briefly review assumptions of network meta-analysis. Then we focus on the statistical analysis of the data: objectives, models (fixed-effects and random-effects), frequentist versus Bayesian approaches, and model validation. A checklist highlights key components of network meta-analysis, and substantial examples illustrate indirect treatment comparisons (both frequentist and Bayesian approaches) and network meta-analysis. A further section discusses eight key areas for future research.

An empirical study of the possible relation of treatment differences to quality scores in controlled randomized clinical trials.

Meta-analytic investigations sometimes use assessments of research quality according to a formal protocol as a tool for improving research synthesis. We asked whether a particular quality scoring system could have a direct use in adjusting the summary estimates of a treatment difference. In an empirical study of the relation of quality scores to treatment differences in published meta-analyses of 7 groups of controlled randomized clinical trials comprising 107 primary studies, we found no relation between treatment difference and overall quality score. We also found no relation between quality score and variation in treatment difference. The level of quality scores has increased at a rate of 9% per decade for three decades, averaging 0.51 on a scale of 0 to 1 for the 1980s, and leaving much room for improvement. Nevertheless, attention to quality of studies by editors, reviewers, and authors may be raising both the level of research done and quality of the reports.

A Single-blind Study of Pulse Oximetry in Children

Oxygen saturation determined by pulse oximetry was monitored in 152 pediatric surgical patients divided into two groups. In one group, the oximeter data and alarms were available (N = 76) to the anesthesia team, and, in the other group, these data were unavailable (N = 76). A trained observer recorded all intraoperative hypoxic episodes and informed the anesthesia team of all major events (i.e., oxygen saturation less than or equal to 85% for greater than or equal to 30 s) (PaO2 approximately 52 mmHg). Thirty-five major events occurred: 24 in the unavailable group, and 11 in the available group (P = 0.021). A greater number of major events occurred in children less than or equal to 2 yr of age (P = 0.013). Hypoxic events diagnosed by the oximeter, but not by the anesthesiologist, were more frequent in the unavailable group (13) than in the available group (5) (P = 0.0495). ASA Physical Status 3 and 4 patients were more likely to suffer a major event (P = 0.009 available, 0.006 unavailable). The pulse oximeter diagnosed hypoxemia before the signs and symptoms of hypoxemia were apparent (i.e., prior to observed cyanosis or bradycardia). Major hypoxic events were unrelated to duration of anesthesia. Major events were evenly distributed among induction, maintenance, and awakening from anesthesia; a greater number of hypoxic events occurred during induction in the unavailable group (P = 0.031). No morbidity was documented in any patient who suffered an hypoxic event.(ABSTRACT TRUNCATED AT 250 WORDS)

Meta-analysis of multiple outcomes by regression with random effects

Earlier work showed how to perform fixed-effects meta-analysis of studies or trials when each provides results on more than one outcome per patient and these multiple outcomes are correlated. That fixed-effects generalized-least-squares approach analyzes the multiple outcomes jointly within a single model, and it can include covariates, such as duration of therapy or quality of trial, that may explain observed heterogeneity of results among the trials. Sometimes the covariates explain all the heterogeneity, and the fixed-effects regression model is appropriate. However, unexplained heterogeneity may often remain, even after taking into account known or suspected covariates. Because fixed-effects models do not make allowance for this remaining unexplained heterogeneity, the potential exists for bias in estimated coefficients, standard errors and p-values. We propose two random-effects approaches for the regression meta-analysis of multiple correlated outcomes. We compare their use with fixed-effects models and with separate-outcomes models in a meta-analysis of periodontal clinical trials. A simulation study shows the advantages of the random-effects approach. These methods also facilitate meta-analysis of trials that compare more than two treatments.

The Sonya Slifka Longitudinal Multiple Sclerosis Study: methods and sample characteristics:

The Sonya Slifka Longitudinal Multiple Sclerosis Study follows a population-based cohort of approximately 2000 people with multiple sclerosis (MS) to study demographic and clinical characteristics, course of illness, utilization and cost of health services, provider characteristics, use of MS specialists and disease modifying agents, and neurologic, economic and psychosocial outcomes. This report describes the study methodology, presents baseline demographic and clinical data, and evaluates the representativeness of the sample. A stratified random sample of persons with established and recently-diagnosed MS selected from the National Multiple Sclerosis Society (NMSS) mailing lists was supplemented with recently-diagnosed patients recruited through systematic nationwide outreach. Baseline data were collected by computer-assisted telephone interviews derived from standardized instruments; data collection continues at six-month intervals. The cohort was comparable to population-based and clinical samples with respect to demographics, course, relapse rate, symptoms, and severity of disability. Almost two-thirds of the cohort needed help with activities of daily living, three-quarters were limited in work or other activities, and half had emotional problems that compromised quality of life. The Slifka Study cohort is broadly representative of the MS population and the database can be used to address questions not answered by natural history studies, clinical databases, or population-based surveys.

Altered Placebo and Drug Labeling Changes the Outcome of Episodic Migraine Attacks

In 459 migraine attacks, information provided to patients (from negative to neutral to positive) modified placebo and medication outcomes in a progressive fashion. Placebo and Medication Effects in Episodic Migraine Placebo and medication effects are intimately related in clinical practice and drug development. In new work, Kam-Hansen et al. investigated how information—ranging from “negative” to “neutral” to “positive”—provided to patients, who received either active drug or placebo, modified their headache pain as measured by patient-reported pain scores. In a randomized order over six consecutive attacks, 66 patients with episodic migraine received either placebo or Maxalt (10-mg rizatriptan) under three information conditions (told placebo, told Maxalt or placebo, told Maxalt). Each participant also reported on an initial no-treatment attack, yielding a total of 459 documented migraine attacks. Maxalt was superior to placebo for pain relief. Increasing information from negative to neutral to positive progressively enhanced the effects of both placebo and Maxalt. The efficacy of open-label placebo was superior to that of no treatment. Relative to no treatment, the placebo, under each information condition, accounted for more than 50% of the drug effect. The benefits of placebo persisted even when the placebo was honestly described. Whether treatment involves medication or placebo, the information provided to patients and the ritual of pill taking are important components of medical care. Information provided to patients is thought to influence placebo and drug effects. In a prospective, within-subjects, repeated-measures study of 66 subjects with episodic migraine, we investigated how variations in medication labeling modified placebo and drug effects. An initial attack with no treatment served as a control. In six subsequent migraine attacks, each participant received either placebo or Maxalt (10-mg rizatriptan) administered under three information conditions ranging from negative to neutral to positive (told placebo, told Maxalt or placebo, told Maxalt) (N = 459 documented attacks). Treatment order was randomized. Maxalt was superior to placebo for pain relief. When participants were given placebo labeled as (i) placebo, (ii) Maxalt or placebo, and (iii) Maxalt, the placebo effect increased progressively. Maxalt had a similar progressive boost when labeled with these three labels. The efficacies of Maxalt labeled as placebo and placebo labeled as Maxalt were similar. The efficacy of open-label placebo was superior to that of no treatment. Relative to no treatment, the placebo, under each information condition, accounted for more than 50% of the drug effect. Increasing “positive” information incrementally boosted the efficacy of both placebo and medication during migraine attacks. The benefits of placebo persisted even if placebo was honestly described. Whether treatment involves medication or placebo, the information provided to patients and the ritual of pill taking are important components of care.

Prevalence of chronic fatigue syndrome in metropolitan, urban, and rural Georgia

BackgroundChronic fatigue syndrome (CFS) is a debilitating illness with no known cause or effective therapy. Population-based epidemiologic data on CFS prevalence are critical to put CFS in a realistic context for public health officials and others responsible for allocating resources.MethodsBased on a random-digit dialing survey we ascertained CFS cases and controls to estimate the prevalence of CFS in metropolitan, urban, and rural populations of Georgia. This report focuses on the 5,623 of 19,381 respondents ages 18 to 59 years old. Fatigued (2,438), randomly selected unwell not fatigued (1,429) and randomly selected well (1,756) respondents completed telephone questionnaires concerning fatigue, other symptoms, and medical history. Subsets of those identified by interview as having CFS-like illness (292), chronic unwellness which was not CFS-like (268 – randomly selected), and well subjects (223, matched to those with CFS-like illness on sex, race, and age) completed a clinical evaluation.ResultsWe estimated that 2.54% of persons 18 to 59 years of age suffered from CFS. There were no significant differences in prevalence of CFS between metropolitan, urban or rural populations or between white and black residents of the three regions. However, there were significant differences in female-to-male ratios of prevalence across the strata (metropolitan female: male 11.2 : 1, urban 1.7 : 1, rural 0.8 : 1).ConclusionWe estimated that 2.54% of the Georgia population suffers from CFS, which is 6- to 10-fold higher than previous population-based estimates in other geographic areas. These differences may reflect broader screening criteria and differences in the application of the case definition. However, we cannot exclude the possibility that CFS prevalence may be higher in Georgia than other areas where it has been measured. Although the study did not identify differences in overall prevalence between metropolitan, urban, and rural Georgia populations, it did suggest the need for additional stratified analyses by geographic strata.

Prediction of First Events of Coronary Heart Disease and Stroke With Consideration of Adiposity

Background— Prediction of coronary heart disease (CHD) and cerebrovascular disease (CeVD) can aid healthcare providers and prevention programs. Previous reports have focused on traditional cardiovascular risk factors; less information has been available on the role of overweight and obesity. Methods and Results— Baseline data from 4780 Framingham Offspring Study adults with up to 24 years of follow-up were used to assess risk for a first CHD event (angina pectoris, myocardial infarction, or cardiac death) alone, first CeVD event (acute brain infarction, transient ischemic attack, and stroke-related death) alone, and CHD and CeVD events combined. Accelerated failure time models were developed for the time of first event to age, sex, cholesterol, high-density lipoprotein cholesterol, diabetes mellitus (DM), systolic blood pressure, smoking status, and body mass index (BMI). Likelihood-ratio tests of statistical significance were used to identify the best-fitting predictive functions. Age, sex, smoking status, systolic blood pressure, ratio of cholesterol to high-density lipoprotein cholesterol, and presence of DM were highly related (P<0.01 for all) to the development of first CHD events, and all of the above except sex and DM were highly related to the first CeVD event. BMI also significantly predicted the occurrence of CHD (P=0.05) and CeVD (P=0.03) in multivariable models adjusting for traditional risk factors. The magnitude of the BMI effect was reduced but remained statistically significant when traditional variables were included in the prediction models. Conclusions— Greater BMI, higher systolic blood pressure, higher ratio of cholesterol to high-density lipoprotein cholesterol, and presence of DM were all predictive of first CHD events, and all but the presence of DM were predictive of first CeVD events. These results suggest that common pathophysiological mechanisms underlie the roles of BMI, DM, and systolic blood pressure as predictors for first CHD and CeVD events.

The effect of nail polish on pulse oximetry.

A randomized, blind study examined the effect of nail polish color on measurement of oxygen saturation by pulse oximetry. Fourteen adult volunteers had blue, green, purple, black, and red nail polish applied to their finger nails. A strip-chart recording of oxygen saturation (Nellcor N100) was made in room air and later interpreted in a blinded fashion. The absorption spectra of the five polishes were determined by spectrophotometry. The spectra of nine other nail polishes and three intravenous dyes also were examined. Black, blue, and green nail polish significantly lowered oximeter readings of oxygen saturation. Blue and green produced greater decreases than purple and red; black produced an intermediate decrease. Some but not all nail polishes absorbed light at the wavelengths used by the pulse oximeter (660 nm and 940 nm). The degree of artifactual desaturation correlated best with the difference between absorbance at 660 nm and absorbance at 940 nm (r = 0.95). Spectrophotometric absorbance data suggest that other colors may interfere with pulse oximetry. On the basis of Spectrophotometric data, brown-red nail polish was predicted to interfere with oximetry; subsequent pulse oximetry measurements confirmed the prediction. Nail polish should be removed routinely before pulse oximetry monitoring.

Effects of information feedback and pulse oximetry on the incidence of anesthesia complications.

No standard outcome measures exist to evaluate the effect of interventions intended to improve the quality of anesthesia care. The authors established a clinically practical definition of outcome, and used it to assess the effect of feedback of information about complications and the effect of pulse oximetry on the rate and severity of important anesthesia-related problems encountered in the operating room (OR) and recovery room (RR). On admission to the RR, the patients anesthetist documented Recovery- Room- Impact E vents (RRIE), defined as an “unanticipated, undesirable, possibly anesthesia-related effect that required intervention, was pertinent to recovery-room care, and did or could cause at least moderate morbidity.” Following a control period with no feedback of data, intense feedback of grouped (anonymous) RRIE rates was provided. Later, pulse oximeters were introduced to all anesthetizing locations. Among 12,088 patients (71% of all RR admissions), 18% had at least one RRIE in the OR or RR. The most common RRIEs were hypotension (4.4%), arrhythmia (3.9%), hypertension (1.5%), intubation difficulties (0.8%), hypoventilation (0.8%), and hypovolemia (0.6%). Feedback of information produced no demonstrable change in the rate of RRIEs. Although significantly fewer patients experienced RRIEs (15.6% vs. 12.4%, P < 0.0001), hypotensive RRIEs (5.2% vs. 3.8%, P = 0.0003), and hypovolemic RRIEs (0.88% vs. 0.42%, P = 0.0017) following the introduction of pulse oximetry in the OR, confounding factors prevent establishment of a cause-and-effect relationship. Quality assurance may require more direct intervention and individual feedback to be effective. Still, the RRIE measure requires minimal effort at low cost and encourages improved transmission of information at the time of admission to recovery-room care.