Systematic review / meta-analysis

From Wikipedia

Contents 1 Methods 1.1 Searching 1.2 Selection 1.3 Validity assessment 1.4 Data abstraction 1.5 Study characteristics 1.6 Quantitative data synthesis 2 Results 2.1 Trial flow 2.2 Study characteristics 2.3 Quantitative data synthesis

Methods

Searching

The overall strategy for information retrieval is explained. The information sources are mentioned in detail (eg, databases, registers, personal files, expert informants, agencies, hand-searching). Also, the conditions or search terms/queries used in information sources (years considered, publication status, language of publication) are also mentioned.

• Systematic literature searches in electronic databases (Cab Abstract, Current Contents, FSTA, and MEDLINE Advanced) were conducted for RCTs of weight reduction and BP by using the search phrase (overweight OR weight reduction OR obes* OR weight OR diet*) AND (hypertension OR blood pressure) AND (trial OR intervention OR random* OR study) as title words or MeSH terms. Article reference lists were examined for additional articles. (Judith E, 2003)

• A broad search of the English-language literature was performed incorporating both electronic and manual components. The electronic search was performed using MEDLINE, Current Contents, and the Cochrane Library databases. MEDLINE (1990-2003, cutoff date June 5, 2003) was searched using the following search terms: obesity/surgery, gastric bypass, gastroplasty, bariatric, gastric banding, "anastomosis, Roux-en-Y," biliopancreatic diversion (including duodenal switch), or jejunoileal bypass. Prior to 1990, the literature offered little to no clinical data on the impact of bariatric surgery weight loss on diabetes, hyperlipidemia, hypertension, and obstructive sleep apnea (Buchwald, 2004).

• A literature search of the MEDLINE database (1966–December 1998), using the Medical Subject Headings hydroxymethyl-glutaryl-CoA reductase inhibitors, simvastatin, lovastatin, pravastatin, coronary disease, and myocardial infarction as well as the key words statin and coronary heart disease was performed. The search was restricted to studies published in English-language journals, conducted in human subjects, and classified as clinical trials in the MEDLINE database. A manual search was also performed using the authors' reference files and reference lists from original communications and review articles.13-15 The contents of 182 abstracts or full-text manuscripts identified during our literature search were reviewed to determine whether they met the criteria for inclusion. Of these abstracts and manuscripts, 29 statin drug treatment trials were identified. Other publications included reviews, letters to the editor, subgroup analysis, and secondary analysis of data from the 29 published trials (LaRosa, 1999).

Selection

Selection of studies based on

• Inclusion and exclusion criteria, that is,
  • defining population,
  • intervention,
  • principal outcomes,
  • study design are mentioned.

• Trials that fulfilled the following criteria were eligible for metaanalysis:(1) an RCT design, (2) published after 1966, (3) conducted in humans, (4) written in English, and (5) nonpharmacologic reduction of body weight. A total of 97 articles was identified, of which 72 were excluded on the basis of the predefined criteria (literature list of excluded trials is available from the authors); ie, they did not fulfill the inclusion criteria (n=21), the intervention period was <8 weeks (n=2), there were co-interventions from which the effect of weight loss could not be separated (n=28), inappropriate or missing BP data (n=5), study objective was not weight reduction (n=4), or overlapping data with other publication(s)(n=12) (Judith E, 2003).

• For inclusion in the subset of studies for data extraction, the screened studies had to report outcomes for one or more of the following comorbid conditions: diabetes, hyperlipidemia, hypertension, and obstructive sleep apnea, or on health care economics. Extracted studies could be of any design, published from 1990 to 2003, and had to have enrolled at least 10 patients undergoing bariatric surgery. All outcomes were preferentially extracted at the time points for which the comorbidity outcomes were available or the latest time point available for follow-up of at least 50% of the population. In studies reporting morbidity improvement, an effort was made to determine both the number of patients evaluated and the time point of evaluation. In some studies, however, the number of patients with a condition at baseline was the only denominator available. Time points for comorbidity improvement were imputed for some studies based on the time point at which other outcomes were reported (principally weight loss). Kin relationships, defined as multiple publications describing the same or overlapping series of patients, were identified and entered into our catalog only once to avoid the double counting of patients (Buchwald, 2004).

• For inclusion, a study had to meet the following criteria: (1) random allocation of study participants to statin or a placebo control group; (2) no intervention difference, other than use of a statin, between the treatment and control groups; (3) intervention duration of at least 4 years; and (4) clinical disease or death as the primary end point. The mean intervention duration in all outcome trials was more than 4 years. Five trials met these criteria and were included in the meta-analysis (LaRosa, 1999).

Validity assessment

• The criteria and process used, including
  • masked conditions
  • quality assessment
  • their findings are included.

Data abstraction

The process or processes used for data abstraction is reported (blidning, reliability checks, completed independently, in duplicate etc). The quality rating scheme and procedures for assessing quality are also mentioned.

• Original articles were retrieved for data abstraction. Two investigators (J.E.N., B.E.S.) independently collected data on population characteristics, study design, and treatment effect on a standard form. For trials in which BP and body weight measurements were recorded at several points in time,data were abstracted for the intervention period during which the maximal BP effect was achieved. If intermediate measurements were not reported, then the total intervention period was used (Judith E, 2003).

• All data were abstracted in duplicate using a standardized protocol and reporting form. Disagreements were resolved by consensus. We did not contact authors to request additional information. Study characteristics recorded were as follows: (1) first author's name, year of publication, and country of origin; (2) number of participants; (3) mean age and age and sex distributions of participants; (4) presence of preexisting myocardial infarction; (5) baseline mean total, LDL-C, high-density lipoprotein cholesterol (HDL-C), and triglyceride levels; (6) net changes in lipids during intervention; (7) type and dosage of statin drug; and (8) intervention duration (LaRosa, 1999).

Study characteristics

This section includes:

• Type of study design
• Participants’ characteristics
• Details of intervention
• Outcome definitions
• How clinical heterogeneity was assessed

• The analysis was based on 25 parallel trials published between 1978 and 2002, comprising 34 strata with a total of 4874 subjects. The duration of intervention until the maximal BP effect (or, if not reported the overall BP effect) was achieved varied between 8 and 260 weeks. Most studies (82%) included both men and women. The mean age of trial populations ranged from 37 to 66 years. Half of the populations were hypertensive (on the basis of initial BP cutoff level of 140/90 mm Hg), and 24% of the populations were taking antihypertensive medication. Mean total duration was 66.6 weeks, and mean duration until the maximal BP effect was achieved was 35.3 weeks. The overall percentage of subjects in RCTs who dropped out after randomization was small (4.8%) (Judith E, 2003).

• After excluding 2 health care economics studies with no weight loss or mortality data, 134 studies were extracted. Fifty-six of the extracted studies were based in North America, 58 in Europe, and 20 were conducted in other locations throughout the world (Australia, New Zealand, South America, Japan, Israel, Saudi Arabia, and Taiwan)Included were 5 randomized controlled trials, 28 nonrandomized controlled trials or series with comparison groups, and 101 uncontrolled case series. The majority of studies were conducted at single centers (n = 126) and only a few were multicenter studies (n = 5). At least 1 categorical outcome of interest (eg, proportion of patients with resolution or improvement in diabetes, hyperlipidemia, hypertension, or obstructive sleep apnea) or 1 continuous outcome of interest (change in a laboratory or physiological measure) was reported by each of the extracted studies(Buchwald, 2004).

• Major coronary events during treatment were abstracted as the primary outcome. They included coronary death, nonfatal myocardial infarction, silent myocardial infarction, or resuscitated cardiac arrest in the Scandinavian Simvastatin Survival Study (4S)1; coronary death or nonfatal myocardial infarction in the West of Scotland Coronary Prevention Study (WOSCOPS),3 the Cholesterol and Recurrent Events Trial (CARE),4-5 and the Long-term Intervention With Pravastatin in Ischaemic Disease (LIPID) trial8; and fatal or nonfatal myocardial infarction, unstable angina, or sudden cardiac death in the Air Force/Texas Coronary Atherosclerosis Prevention Study (AFCAPS/TexCAPS)(LaRosa, 1999).

Quantitative data synthesis

• Principal measures of effect (eg, relative risk)
• Method of combining results - statistical testing and confidence intervals
• Handling of missing data
• How statistical heterogeneity was assessed
• A rationale for any a-priori sensitivity and subgroup analyses
• Any assessment of publication bias

• Changes in BP from baseline in the control group were subtracted from changes in the intervention group to obtain the net BP effect of weight reduction. SEs of the treatment effect were also abstracted. If not reported, SEs were derived from SDs or SEs for BP effects within groups, confidence intervals (CIs), test statistics, or probability values. A random-effects model was chosen for meta-analysis to account for heterogeneity in BP effect among trials. An advanced statistical technique for meta-analysis of continuous outcomes was applied that takes into account both within- and between-study variation (Judith E, 2003).

• A random effects model was used in the meta-analysis. The mean (95% confidence interval) percentage of excess weight loss was 61.2% (58.1%-64.4%) for all patients; 47.5% (40.7%-54.2%) for patients who underwent gastric banding; 61.6% (56.7%-66.5%), gastric bypass; 68.2% (61.5%-74.8%), gastroplasty; and 70.1% (66.3%-73.9%), biliopancreatic diversion or duodenal switch. Operative mortality (≤30 days) in the extracted studies was 0.1% for the purely restrictive procedures, 0.5% for gastric bypass, and 1.1% for biliopancreatic diversion or duodenal switch. Diabetes was completely resolved in 76.8% of patients and resolved or improved in 86.0%. Hyperlipidemia improved in 70% or more of patients. Hypertension was resolved in 61.7% of patients and resolved or improved in 78.5%. Obstructive sleep apnea was resolved in 85.7% of patients and was resolved or improved in 83.6% of patients (Buchwald, 2004).

• Both proportional (1 - odds ratio [OR]) and absolute risk reduction were used to measure the effect of statin drug treatment on clinical outcomes. The numbers of various outcomes for both the statin and placebo groups were recorded for each study using 2 x 2 tables. The Peto method was used to calculate pooled ORs of outcomes associated with statin therapy.40 For each trial, the number of individuals in the treatment group in whom an end point of interest was observed (O) was compared with the number that would, if treatment had no effect, have been expected (E) on the basis of the overall experience in the treatment and control groups combined (LaRosa, 1999).

Results

Trial flow

• Provide a meta-analysis profile summarising trial flow
• Flowchart

Study characteristics

• Presents descriptive data for each trial
  • Age
  • Sample size
  • Intervention
  • Dose
  • Duration
  • Follow-up period

• The analysis was based on 25 parallel trials published between 1978 and 2002, comprising 34 strata with a total of 4874 subjects. The duration of intervention until the maximal BP effect (or, if not reported the overall BP effect) was achieved varied between 8 and 260 weeks. Most studies (82%) included both men and women. The mean age of trial populations ranged from 37 to 66 years. Half of the populations were hypertensive (on the basis of initial BP cutoff level of 140/90 mm Hg), and 24% of the populations were taking antihypertensive medication. Mean total duration was 66.6 weeks, and mean duration until the maximal BP effect was achieved was 35.3 weeks. The overall percentage of subjects in RCTs who dropped out after randomization was small (4.8%) (Judith E, 2003).

• A total of 136 fully extracted studies, which included 91 overlapping patient populations (kin studies), were included for a total of 22 094 patients. Nineteen percent of the patients were men and 72.6% were women, with a mean age of 39 years (range, 16-64 years). Sex was not reported for 1537 patients (8%). The baseline mean body mass index for 16 944 patients was 46.9 (range, 32.3-68.8) (Buchwald, 2004).

• A total of 30,817 participants were included in these trials. Mean follow-up time was 5.4 years and mean age was 59 years. Only men younger than 65 years were included in WOSCOPS, while women and participants who were aged 65 years or older were included in the remaining 4 trials. Three trials (4S, CARE, and LIPID) were conducted in patients with a history of coronary heart disease (secondary prevention trials) and 2 trials (WOSCOPS and AFCAPS/TexCAPS) were conducted in a healthy population (primary prevention trials). Three trials (4S, CARE, and LIPID) were conducted in patients with a history of coronary heart disease (secondary prevention trials) and 2 trials (WOSCOPS and AFCAPS/TexCAPS) were conducted in a healthy population (primary prevention trials). (LaRosa, 1999).

Quantitative data synthesis

The methods for estimating variance for individual studies, model to be used, reason for using that model, approach to missing data, how quality rating shceme will be used in the analysis are all stated in this section.

• Reports agreement on the selection and validity assessment
• Presents simple summary results
  • for each treatment group in each trial
  • for each primary outcome
  • Presents data needed to calculate effect sizes and confidence intervals in intention-to-treat analyses (eg 2x2 tables of counts, means and SDs, proportions)

• Analyses were performed only on the data from the studies in the data extraction subset. Study, patient, and treatment-level data were summarized using basic descriptive statistics (simple counts and means). The number of patients enrolled or randomized was used in the calculation of study and patient demographics. Efficacy outcomes of interest were synthesized via meta-analytic pooling of similar surgery group results across studies with stratification by the type of surgery. In addition, meta-analysis of within-study surgery effects on weight loss and diabetes-related outcomes were stratified by studies with extractable outcomes for a general population compared with subgroups of patients with diabetes or impaired glucose tolerance. Meta-analyses of all efficacy outcomes were conducted using a random-effects model, estimated by using the restricted maximum likelihood method.21-22 Efficacy outcomes included both proportions (eg, response rates) and raw mean before and after changes (eg, absolute weight changes). The random-effects model meta-analyses take into account both study sample size and the estimate of between-study variation (ie, study heterogeneity) when weighting study effects. Meta-analytic means and mean changes are expressed with 95% confidence intervals (CIs) (Buchwald, 2004).

• The mean reduction (weighted by sample size) in total cholesterol, LDL-C, and triglyceride levels was -20%, -28%, and -13%, respectively, and HDL-C was increased by an average of 5% among the 5 trials. When the results from the 5 trials were pooled, a significant reduction in the odds of major coronary events and coronary deaths (P<.001 for both) was observed among the participants allocated to active treatment. Active treatment was associated with a 34% risk reduction (95% CI, 23%-43%; P<.001) in major coronary events in the 2 primary prevention trials and a 30% risk reduction (95% CI, 24%-35%; P<.001) in the 3 secondary prevention trials. In all trials, the odds of coronary events were reduced for those assigned to active treatment compared with controls, and the risk reduction was statistically significant in 2 of 4 trials in women and all 5 trials in men (LaRosa, 1999).