Carskadon, M.A., & Dement, W.C. (2011). Monitoring and staging human sleep. In M.H. Kryger, T. Roth, & W.C. Dement (Eds.), Principles and practice of sleep medicine, 5th edition, (pp 16-26). St. Louis: Elsevier Saunders. Chapter 2 – Normal Human Sleep : An Overview Mary A. Carskadon, William C. Dement Abstract
Eprints.keele.ac.ukComparative cardiovascular effects of thiazolidinediones:systematic review and meta-analysis of observationalstudies Yoon Kong Loke, senior lecturer in clinical pharmacology,1 Chun Shing Kwok, medical student,1 Sonal Singh,assistant professor of medicine2 1School of Medicine, Health Policy because of concerns about adverse cardiovascular and Practice, University of East Objective To determine the comparative effects of the events.2 Rosiglitazone and pioglitazone are the available Anglia, Norwich NR4 7TJ, UK thiazolidinediones (rosiglitazone and pioglitazone) on thiazolidinediones in North America, but meta-analyses Johns Hopkins University School myocardial infarction, congestive heart failure, and of Medicine, Baltimore, MD of randomised controlled trials have suggested an mortality in patients with type 2 diabetes.
increased risk of ischaemic cardiovascular events with Correspondence to: Y K Loke Design Systematic review and meta-analysis of rosiglitazone.34 In contrast, meta-analysis of trials of pio- glitazone indicates the possibility of an ischaemic cardio- Data sources Searches of Medline and Embase in Cite this as: BMJ 2011;342:d1309 vascular benefit.5 Robust evidence also shows that both September 2010.
drugs increase the risk of congestive heart failure and Study selection Observational studies that directly fractures, but whether any meaningful difference exists compared the risk of cardiovascular outcomes for in the magnitude of risk between the two thiazolidine- rosiglitazone and pioglitazone among patients with type diones is not known.67 The European Medicines 2 diabetes mellitus were included.
Agency has recommended the suspension of marketing Data extraction Random effects meta-analysis (inverse authorisation for rosiglitazone, whereas the US Food variance method) was used to calculate the odds ratios and Drug Administration has allowed the continued for cardiovascular outcomes with thiazolidinedione use.
marketing of rosiglitazone with additional restrictions.8 The I2 statistic was used to assess statistical No long term trials with cardiovascular outcomes have directly compared these two drugs. Clinical trials Results Cardiovascular outcomes from 16 observational have strict selection criteria that may exclude partici- studies (4 case-control studies and 12 retrospective pants at high risk of adverse events, and adverse cohort studies), including 810 000 thiazolidinedione cardiovascular outcomes can be rare in such trials.9 users, were evaluated after a detailed review of 189 On the other hand, population based observational citations. Compared with pioglitazone, use of studies resemble clinical practice, where patients may rosiglitazone was associated with a statistically have risk factors for cardiovascular disease or comor- significant increase in the odds of myocardial infarction bidities. Therefore, consideration of the evidence from (n=15 studies; odds ratio 1.16, 95% confidence interval carefully conducted observational studies is essential 1.07 to 1.24; P<0.001; I2=46%), congestive heart failure to determine if any difference in cardiovascular events (n=8; 1.22, 1.14 to 1.31; P<0.001; I2=37%), and death or mortality exists between the two drugs.
(n=8; 1.14, 1.09 to 1.20; P<0.001; I2=0%). Numbers Our objective was to systematically determine the needed to treat to harm (NNH), depending on the comparative effects of rosiglitazone and pioglitazone population at risk, suggest 170 excess myocardial on cardiovascular outcomes (myocardial infarction infarctions, 649 excess cases of heart failure, and 431 and congestive heart failure) and mortality from obser- excess deaths for every 100 000 patients who receive vational studies in patients with type 2 diabetes. We rosiglitazone rather than pioglitazone.
aimed specifically to calculate the pooled odds ratios Conclusion Among patients with type 2 diabetes, use of for adverse cardiovascular events with rosiglitazone rosiglitazone is associated with significantly higher odds compared with pioglitazone—that is, the relative like- of congestive heart failure, myocardial infarction, and lihood of cardiovascular harm if rosiglitazone was used death relative to pioglitazone in real world settings.
rather than pioglitazone.
Troglitazone, the first thiazolidinedione, was withdrawn Eligibility criteria from the market because of liver toxicity.1 Muraglitazar, We selected controlled observational (non-randomised) a dual peroxisome proliferator activated receptor studies that reported on cardiac outcomes in patients (PPAR) agonist, failed to achieve regulatory approval receiving rosiglitazone compared with pioglitazone.
BMJ ONLINE FIRST bmj.com Any uncertainties or discrepancies between the two Articles retrieved from Articles retrieved by reviewers were resolved through consensus after re- PubMed and Embase checking titles in GSK study register (n=4) checking of the source data and consultation with thethird reviewer. We also contacted authors if any areasof uncertainty needed clarification.
Titles and abstracts screened for studies Where different timings and durations of thiazolidi- that might be potentially relevant (n=189) nedione use were reported in the study participants, we Review articles, trial reports, or clearly not observational pre-specified that data would be preferentially extra- studies of cardiovascular events with rosiglitazone v cted from the participants with current or most recent pioglitazone in patients with type 2 diabetes (n=153) use, until cessation of treatment. We also aimed to Potentially relevant studies for full checking (n=36) extract risk estimates pertaining to overall use in theentire cohort rather than in any specific subgroups.
Excluded (n=20): Did not report on direct comparison of rosiglitazone v pioglitazone (n=15) Earlier study that used same database population as a later study (n=1) In accordance with the recommendations of the Had duplicate data (n=4) Cochrane Adverse Effects Methods Group, wechecked the methods of selection of participants Observational studies with direct comparison of cardiac outcomes or mortality for rosiglitazone v pioglitazone (n=16) (including baseline characteristics and adjustment forconfounders), nature of follow-up, ascertainment ofdrug use, and definition and monitoring of adverse Fig 1 Flow diagram of process of selection of articles for outcomes.10 To counter selective reporting bias, we contacted authors when relevant cardiovascular out-comes were potentially measured but were not We included studies of a cohort or case-control design reported or were stated to be non-significant. We that enrolled participants with type 2 diabetes mellitus.
used a funnel plot to assess publication bias.
The primary outcome of interest was myocardial infarc-tion. Secondary outcome measures were congestive heart failure and overall mortality. Eligible studies had We used RevMan 5.0.25 (Nordic Cochrane Centre) to to present one of the following: odds ratio, relative risk, do random effects meta-analysis using the inverse var- hazard ratio, or sufficient raw data to enable calculationof the odds ratio where not otherwise reported.
iance method for pooled odds ratios. We used the fixedeffects model for sensitivity analysis. We assumed simi- larity between the odds ratio and other relative mea-sures such as relative risk, rate ratios, or hazard ratios We searched Medline and Embase by using Ovid SP because cardiovascular events and deaths were rare (from inception to the end of September 2010), with the search terms (pioglitazone or rosiglitazone or thia-zolidinedione$).mp and (myocardial-infarction or Where possible, we aimed to pool adjusted odds cardiovascular or cardiac or heart).mp and (cohort or ratios from the primary studies; otherwise, we used case-control or observational or retrospective).mp. We raw outcome data to yield unadjusted odds ratios. In did not use any language restrictions, but we limited view of the potential diversity of study designs, we the search to human studies. Additionally, we signed grouped the studies for the analysis according to stu- up with PubMed to receive automated electronic noti- dies for which only the unadjusted odds ratios were fication of any new articles containing the above search available, with no correction for baseline differences terms. To identify unpublished studies, we reviewed or confounding, and those for which we were able to the regulatory authorities websites (US Food and extract odds ratios adjusted for potential confounders.
Drug Administration and European Medicines For consistency in direction of risk comparisons, we Agency), as well as the study registers of the drug man- used the odds ratio to assess the magnitude of risk for ufacturers GlaxoSmithKline and Takeda. We checked rosiglitazone use compared with that for pioglitazone the bibliographies of included studies and recent use. For studies that reported the odds ratio for piogli- review articles for additional relevant articles.
tazone compared with rosiglitazone, we used the reci-procal of the point estimate and the bounds of the Study selection and data extraction Two reviewers (CSK and YKL or SS) checked all titles We estimated the number needed to treat to harm and abstracts for studies that could potentially meet the per year (NNH) (and 95% confidence interval) by inclusion criteria. We retrieved full reports of these applying the pooled odds ratio from the meta-analysis potentially eligible studies for detailed assessment by to the annual rate of the event in different two reviewers (CSK and YKL), who then indepen- populations.12 The NNH is the number of patients dently extracted information on study design, drug with type 2 diabetes who need to be treated with rosi- use, study location, characteristics of participants, and glitazone rather than pioglitazone for one additional relevant outcomes on to a preformatted spreadsheet.
patient to have an adverse outcome.
BMJ ONLINE FIRST bmj.com Table 1 Design and characteristics of included studies Study design and data source Selection criteria Risk estimates (95% CI) Retrospective cohort study (TRIAD); 564 rosiglitazone; Type 2 diabetes, age >18 years, not pregnant, community patients; USA, 1999 to 334 pioglitazone community dwelling, English or Spanish speaking, and thiazolidinediones were (in health plans in which enrolled in health plan for ≥18 months; excluded available on formulary: both thiazolidinediones if >1 type of thiazolidinedione prescription MI HR 1.3 (0.31 to 5.37); mortality 0.69 (0.28 to 1.69) Retrospective cohort study in 1879 rosiglitazone; Aged >18 years with diabetes or HbA1C >6% and Based on entire cohort with Partners Healthcare System ≥1 oral diabetes drug; excluded if used metformin adjustment for known risk covering hospital and community or thiazolidinedione for polycystic ovaries factors: MI RR 1.7 (1.1 to 2.6) patients; USA, January 2000 to July2006 Nested case-control study; Acute MI: 2244 cases and Previous metformin users; excluded if received other Based on overall use: hospital and community patients; 8903 controls; drug use: oral antidiabetic drug or insulin within 365 days before MI OR 1.00 (0.67 to 1.49) Pharmanet database, BC, Canada, 462 rosiglitazone and starting metformin or emigrated/died before May 2003 May 2003 to March 2007 Graham 201017 New user inception cohort; 67 593 rosiglitazone; ≥6 month enrolment and >65 years who started MI HR 1.06 (0.96 to 1.18); community patients; Medicare, 159 978 pioglitazone rosiglitazone or pioglitazone; excluded if residing HF 1.25 (1.16 to 1.34); USA, July 2006 to June 2009 in a hospital, long term care home, or hospice mortality 1.14 (1.05 to 1.24) Retrospective cohort study; NHI 49 624 rosiglitazone; Newly diagnosed type 2 diabetes with ≥3 prescriptions Unadjusted data for entire claims database, Taiwan, 2000 to 12 010 pioglitazone of oral diabetes drug; excluded if had type 1 diabetes or cohort: MI ros 1984/49 624, had been on insulin during study period pio 356/12 010; HF ros 664/49 624, pio 115/12 010 Retrospective new user cohort 16 951 rosiglitazone; Residents of Ontario, >66 years of age starting MI HR 1.05 (0.90 to 1.23); HF study; Ontario Public Drug Benefit 22 785 pioglitazone.
thiazolidinedione treatment; excluded if using insulin 1.30 (1.15 to 1.45); mortality Program, Canada, 2002 to 2008 1.16 (1.02 to 1.33) Nested case-control study within MI: 9870 cases and Patients with type 2 diabetes and ≥1 prescription claim MI OR 1.12 (0.99 to 1.26) diabetes cohort of Integrated 29 610 controls; drug use: for antidiabetic drug, with ≥1 year enrolment in HealthCare Information Services 3839 rosiglitazone and healthcare plan; those with heart failure or ischaemic claims database; USA, 1999 to 3343 pioglitazone heart disease were included, but those with MI were Nested case-control study; 1886 current rosiglitazone Aged ≥66 years with diabetes and dispensed ≥1 oral Unadjusted data current community patients; Ontario, users; 929 current hypoglycaemic agent in study period; excluded if users: MI ros 335/1886, pio Canada, 2002 to 2005 pioglitazone users received insulin in year preceding cohort entry 134/920; HF ros 426/1907,pio 160/929; mortality ros434/1716 pio 165/715 Retrospective cohort study; 7282 rosiglitazone; Patients with two records of diabetes between 2002 Full cohort: MI or CAD HR 1.0 community patients in THIN GP 2244 pioglitazone and 2006 and ≥40 years old database; UK, 2002 to 2006 Retrospective cohort study; 1079 rosiglitazone; Type 2 diabetes with prescription for rosiglitazone, HF HR 0.84 (0.52 to 1.35); Cleveland Clinic Electronic Health 1508 pioglitazone pioglitazone, metformin, or sulfonylurea, age >18 years mortality 1.23 (0.79 to 1.92) Records; USA, October 1998 to with no history of dialysis, CAD, or HF; excluded if prescribed insulin or multiple oral agents Nested case-control study; claims MI: 1681 cases and Patients aged 18-84 years with a filled prescription for MI OR 1.26 (0.79 to 2.00) database of Prescription Solutions 6653 controls; drug use: antidiabetic drug or exenatide during study period; cohort in 5 states in USA, January 1039 rosiglitazone and excluded if had type 1 diabetes, cancer, renal or liver 2000 to June 2006 failure, organ transplantation, or HIV infection Retrospective cohort study; 140 082 rosiglitazone; Patients aged 35-90 years with episode of care MI HR 1.34 (0.86 to 2.09); community patients; UK General 45 807 pioglitazone between 1990 and 2005 associated with clinical or HF 1.04 (0.75 to 1.44); Practice Research Database, referred event for diabetes; excluded if date of death mortality 1.36 (1.05 to 1.76) January 1990 to December 2005 Retrospective cohort study; 57 000 rosiglitazone; Users of oral hypoglycaemic agents who had On treatment summary: pharmacy and medical claims 51 000 pioglitazone ≥6 months' membership in health plan, age >18 years; MI 1.21 (0.95 to 1.54) database (Pharmetrics) covering excluded if in health plans for which data had been >80 health plans; USA, 2000 to previously used in similar studies Retrospective cohort study; 18 319 rosiglitazone; Aged >18 years with new rosiglitazone or pioglitazone MI HR 0.94 (0.75 to 1.18); HF medical/pharmacy claims in 18 309 pioglitazone claim; excluded if not in health plan >365 days before 1.10 (0.94 to 1.31); mortality WellPoint database; USA, January index date or had pre-index pharmacy claim of insulin; 1.02 (0.86 to 1.21) 2001 to December 2005 those with previous cardiovascular events were notexcluded New user cohort study; Medicare 14 101 rosiglitazone; Age >65 years with new prescription for On-drug analysis: MI IRR 1.08 database in New Jersey and 14 260 pioglitazone thiazolidinedione; excluded if used troglitazone or (0.93 to 1.25); HF 1.13 (1.01 Pennsylvania, USA, January 2000 fixed dose combination with metformin to 1.26); mortality 1.15 (1.05 Ziyadeh 200929 Retrospective cohort study; i3 47 501 rosiglitazone; Age >18 years, starting rosiglitazone or pioglitazone, Based on censoring at proprietary research database of 47 501 pioglitazone followed by >6 months of health plan membership; discontinuation of treatment medical claims in USA, July 2000 to troglitazone users excluded (regimen stop): MI HR 1.41 BMJ ONLINE FIRST bmj.com of patients were male across 15 studies. Table 1 shows We used the I2 statistic to assess statistical heterogene- the main characteristics of the studies and participants; ity. I2 values of 30-60% represented a moderate level of table 2 shows the outcomes, interventions, and quality Figure 1 shows the process of selection of studies. We The included studies were broadly similar in terms of retrieved 16 observational studies involving 810 000 ascertainment of drug use and cardiovascular out- thiazolidinedione users (429 000 patients taking rosi- comes (table 2); they relied mainly on computerised glitazone and 381 000 taking pioglitazone), after a diagnostic codes, pharmacy claims databases, and ret- detailed review of 189 citations.14-29 Fifteen studies rospective chart reviews. Few researchers made reported on the outcome of myocardial infarction, attempts to verify drug history directly with the eight studies reported on the outcome of congestive patients or to check the validity of the prescriptions heart failure, and eight studies reported on mortality.
data source; only one study was deemed to have spe- We found 12 retrospective cohort studies and four cific validation of drug use.14 Most studies reported the case-control studies. The mean or median follow-up accuracy of outcome ascertainment on the basis of his- time ranged from 105 days to 7.1 years. Four studies torical validation studies; only two studies specifically reported duration of thiazolidinedione use, with a cross checked or validated outcomes for this range of 215 to 450 days. The mean age of participants analysis.15 23 Both of these studies showed a limited ranged from 54 to 76 years across studies, but most degree of misclassification.15 23 study participants were generally aged above None of the studies provided details about the sever- 60 years; only two studies reported the average age of ity and consequences of the cardiac adverse events.
their participants as under 60 years. An average of 55% Almost all the studies used a wide variety of variablesto adjust for potential confounders. Two cohort studies Study or subgroup checked specifically for similarities between the rosigli- tazone and pioglitazone populations and did not find Adjusted odds ratio 1.30 (0.31 to 5.37) evaluated.17 19 We were able to use adjusted risk esti- Brownstein 201015 1.70 (1.10 to 2.60) mates for most studies, except for two studies for 1.00 (0.67 to 1.49) which we calculated odds ratios from the raw 1.06 (0.96 to 1.18) 1.05 (0.90 to 1.23) 1.12 (0.99 to 1.26) Myocardial infarction 1.00 (0.80 to 1.30) Compared with pioglitazone, use of rosiglitazone was 1.26 (0.79 to 2.00) 1.34 (0.86 to 2.09) associated with a significantly increased odds of myo- 1.21 (0.95 to 1.54) cardial infarction from 15 studies (pooled odds ratio 0.94 (0.75 to 1.18) 1.16, 95% confidence interval 1.07 to 1.24; P<0.001) Winkelmayer 200828 1.08 (0.93 to 1.25) (fig 2). We found a moderate level of heterogeneity 1.41 (1.13 to 1.75) (I2=46%) for the pooled results for myocardial infarc- Subtotal (95% CI) 1.11 (1.04 to 1.18) tion, which stemmed from combining the unadjusted Test for heterogeneity: τ2=0.00, and adjusted studies together for the overall estimate.
χ2=14.25, df=12, P=0.28, I2=16%Test for overall effect: z=3.19, P=0.001 Heart failureBased on the pooled results of eight studies, the odds of Unadjusted odds ratio congestive heart failure were statistically significantly 1.36 (1.22 to 1.53) higher for rosiglitazone than for pioglitazone (odds Lipscombe 200721 1.27 (1.02 to 1.58) ratio 1.22, 1.14 to 1.31; P<0.001), with moderate statis- Subtotal (95% CI) 1.34 (1.21 to 1.48) tical heterogeneity (I2=37%) (fig 3).
Test for heterogeneity: τ2=0.00, χ2=0.34, df=1, P=0.56, I2=0% Overall mortality Test for overall effect: z=5.68, P<0.001 The odds of death were statistically significantly higherfor rosiglitazone than for pioglitazone when we pooled 1.16 (1.07 to 1.24) eight studies, with an odds ratio of 1.14 (1.09 to 1.20; Test for heterogeneity: τ2=0.01, P<0.001) (fig 4). We found no evidence of statistical χ2=25.89, df=14, P=0.03, I2=46% heterogeneity for this outcome (I2=0%).
Test for overall effect: z=3.87, P<0.001 Number needed to treat for harm In a low risk population (age 45-64 years) with type 2 Fig 2 Meta-analysis of odds ratio for myocardial infarction with rosiglitazone versus diabetes but no previous history of myocardial infarc- tion, the underlying incidence of myocardial infarction BMJ ONLINE FIRST bmj.com Study or subgroup we did a post hoc analysis by excluding the single study that had a substantial proportion of patients recruited Adjusted odds ratio after May 2007.17 This did not appreciably change the 1.25 (1.16 to 1.34) direction and magnitude of the estimates for myo- 1.30 (1.15 to 1.45) cardial infarction (odds ratio 1.17, 1.08 to 1.27), heart Pantalone 200923 0.84 (0.52 to 1.35) failure (1.21, 1.10 to 1.33), and overall mortality (1.13, 1.04 ( 0.75 to 1.44) 1.04 to 1.24). Further exclusion of another study 1.10 (0.94 to 1.31) (recruitment dates 2002 to 2008) did not appreciably Winkelmayer 200828 1.13 ( 1.01 to 1.26) change the odds ratios for myocardial infarction (1.19, Subtotal (95% CI) 1.19 (1.10 to 1.28) 1.09 to 1.29), heart failure (1.18, 1.06 to 1.33), and over- Test for heterogeneity: τ2=0.00, all mortality (1.14, 1.06 to 1.22).19 χ2=7.88, df=5, P=0.16, I2=37%Test for overall effect: z=4.58, P<0.001 Assessment of publication bias Unadjusted odds ratio The funnel plot showed that risk estimates stemmed 1.40 (1.15 to 1.71) mostly from large, precise studies that seemed to be Lipscombe 200721 1.38 (1.13 to 1.69) fairly well distributed, with no definite evidence of Subtotal (95% CI) 1.39 (1.21 to 1.60) asymmetry (fig 5).
Test for heterogeneity: τ2=0.00, χ2=0.01, df=1, P=0.91, I2=0% Test for overall effect: z=4.58, P<0.001 Our results suggest a modest but statistically significantincrease in the odds of myocardial infarction (approxi- 1.22 (1.14 to 1.31) mately 16%), congestive heart failure (approximately Test for heterogeneity: τ2=0.00, 23%), and mortality (approximately 14%) with use of χ2=11.20, df=7, P=0.13, I2=37% rosiglitazone compared with those for pioglitazone use Test for overall effect: z=5.64, P<0.001 in real world studies among patients with type 2 diabetes.
The consistency in the magnitude of increased risk for the different cardiac outcomes, as well as mortality, indi- cates that this is unlikely to be a chance finding. Other Fig 3 Meta-analysis of odds ratio for heart failure with rosiglitazone versus pioglitazone strengths of our analysis include the large number ofthiazolidinedione users (around 810 000) and the was 1.08% per year.30 Use of rosiglitazone here would absence of substantial statistical heterogeneity, which result in an annual NNH of 587 (95% confidence inter- suggests that the risk is maintained across most popula- val 392 to 1339). This can be equated to 170 excess tions and is unaffected by geographical variations.
myocardial infarctions for every 100 000 patientswho received rosiglitazone rather than pioglitazone.
Comparison with other studies In a US cohort of patients with type 2 diabetes, the Our synthesis of evidence from observational studies baseline incidence of heart failure was found to be extends the findings of a cardiovascular hazard with 3.08% per year.31 Use of rosiglitazone here would rosiglitazone from meta-analysis of clinical trials to result in an annual NNH of 154 (110 to 241). This real world settings and suggests the possibility of a can be equated to 649 excess cases of heart failure for cardiovascular difference between the two drugs.3-5Adjusted indirect comparisons of the risk estimates every 100 000 patients who received rosiglitazone from meta-analysis of myocardial infarction and heart rather than pioglitazone.
failure in randomised controlled trials shows that rosi- In a large French registry study of patients with type glitazone is associated with an increased relative risk of 2 diabetes and atherosclerosis, the underlying mortal- 1.58 (95% confidence interval 1.14 to 2.20) for myo- ity rate was found to be 3.15% per year.32 Use of rosi- cardial infarction and 1.48 (1.01 to 2.18) for heart fail- glitazone here would result in an annual NNH of 232 ure, compared with pioglitazone.3-534 The direction of (163 to 360). This can be equated to 431 excess deaths effect for both outcomes is consistent with our analysis, for every 100 000 patients who received rosiglitazone whereas the relatively lower point estimates seen in our rather than pioglitazone.
analysis may reflect the generally more conservativenature of estimates of harm that has been noted with Sensitivity analysis observational studies.35 Participants in trials may differ Meta-analysis using the fixed effects model yielded from those in observational studies, because most of estimates that were similar in direction and magnitude the observational studies recruited a wider, more gen- to those from the random effects model for myocardial eralisable range of patients by not enforcing rigid inclu- infarction (odds ratio 1.15, 1.10 to 1.21), heart failure sion and exclusion criteria relating to comorbid (1.23, 1.17 to 1.29), and overall mortality (1.14, 1.09 to conditions. Of the 16 included studies, two excluded patients with existing cardiac conditions (so that they In view of potential patient selection bias arising could study incident disease)20 23 and only one after publication of a meta-analysis in May 2007 that excluded patients who had comorbidities such as showed increased myocardial risk with rosiglitazone,33 renal or liver disease.
BMJ ONLINE FIRST bmj.com Table 2 Drug use, study outcomes, and potential sources of bias Ascertainment of drug use Ascertainment of outcomes Adjustment for confounders Ascertained by health plan records MI and all cause mortality ascertained by ICD codes on Age, sex, race, income, history of diabetic nephropathy, history of for prescriptions filled; average of health plan administrative data and national death cardiovascular disease, insulin use, and health plan Based on electronic records and MI ascertainment based on ICD codes and randomly Age, sex, cardiovascular disease, antihypertensive and lipid lowering randomly checked case notes/ selected case notes/discharge summaries with drugs, and Charlson score discharge summaries, with 94% sensitivity of 94% and specificity of 74% for outcomes sensitivity and specificity for druguse Pharmanet database of MI ascertainment based on ICD codes of hospital Duration of diabetes; congestive heart failure; angiography; prescriptions dispensed at admission records (primary reason for admission) revascularisation; ischaemic stroke; TIA; previous MI; angina; renal community pharmacies (data collected by Ministry of Health; controls were patients disease; Romano comorbidity score; use of cardiac drugs, clopidogrel, and quality checks done by Pharmanet) starting metformin matched on age, sex, number of insulin; and past use of metformin, glitazones, and sulfonylureas family members, health plan enrolment, and income Drug claims linked to Medicare MI and heart failure based on ICD discharge codes with Sex, age, race, low income, extended care, Charlson score, cardiovascular database for prescription drugs positive predictive values of >90%; mortality disease and drugs, lipid lowering drugs, and comorbidities from January 2006; median follow- ascertained by linkage to social security master beneficiary record database, which captures 95% ofdeaths for older people Mean use of around 450 days MI and heart failure ascertainment based on ICD codes Data used in unadjusted form in meta-analysis ascertained with prescription claims for inpatient claims Follow-up median of 292 days for All cause mortality, MI, and heart failure obtained from Age, sex, residence, socioeconomic status, year of entry, duration of rosiglitazone and 294 days for national ambulatory care reporting system database, diabetes, acute MI, angina, congestive heart failure, coronary angiography, pioglitazone; ascertainment based Canadian Institute for Health information discharge CABG, PCI, Charlson index, history of renal disease, and previous drugs on computerised prescription database, and Ontario health insurance database (antihypertensives, aspirin, NSAIDs, nitroglycerin preparations, statins, oral hypoglycaemics, digoxin) Follow-up mean of 2.1 years with Cases had ICD code for hospital admission for MI Age; use of ACE inhibitors, β blockers, diuretic, or nitrate; hyperlipidaemia; drug use window of 3 months, occurring ≥3 months after diagnosis of diabetes; hypertension; and CAD inferred from prescription claims controls were randomly selected from eligible matchedpatients within cohort who did not have ICD code for MI Unclear ascertainment; median MI, heart failure, and mortality data from registered Data used in unadjusted form in meta-analysis follow-up 3.8 years persons databases and hospital discharge summaryabstract database Thiazolidinedione use on average MI and coronary artery disease (MI, unstable angina, Age; sex; BMI; HbA1C; smoking; chronic kidney disease; eGFR; mean 3.5 years; data based on cardiac death, coronary artery reperfusion procedure) arterial blood pressure; and history of MI, unstable angina, or cardiac computerised prescription records based on computerised read codes in general practice proceduredatabase; approximately 3% lost to follow-up Pantalone 200923 Drug use at baseline based on Heart failure and mortality data from ICD codes and Age, sex, race, eGFR, albumin/urine creatinine ratio, HbA1C, BMI, systolic database information at single electronic health records database; small proportion of blood pressure, diastolic blood pressure, HDL, LDL, triglycerides, smoking healthcare centre mortality records from social security death index had status, cardiovascular drugs, new diabetes, and median household errors on cross checking and were corrected for Drug use based on prescription MI cases based on ICD codes for hospital admission; Age, cardiovascular risk score, non-cardiovascular acute hospital claims from pharmacy database controls were matched on various parameters with admission, COPD, and use of oestrogen therapy specific group constructed for analysis of rosiglitazonev pioglitazone Drug use based on database MI, heart failure, and mortality; unclear outcome Age, sex, duration of diabetes, complications of diabetes, cardiovascular information; median follow-up and peripheral artery disease, co-prescribed drugs, BMI, cholesterol concentration, systolic blood pressure, HbA1C, creatinine concentration,albumin concentration, and smoking status Drug use from PharMetrics MI based on hospital discharge diagnosis and ICD code Propensity score used to adjust according to demographics, calendar time, database; mean on-treatment time use of antidiabetic drugs, history of MI, coronary revascularisation, angina, 8 months and overall follow-up ACS, congestive heart failure, hyperlipidaemia, hypertension, obesity, ranged from 12 to 18 months smoking, use of cardiovascular drugs Drugs from pharmacy records; MI, heart failure, and mortality from medical claims, ICD Propensity score used to adjust for age, sex, health plan, Deyo-Charlson unclear ascertainment; mean codes, and national death index plus database; mean comorbidity index score, cardiovascular and peripheral vascular disease, duration of treatment 14.6 months follow-up 19.6 months cardiovascular and antidiabetic drugs, obesity, smoking status, anddiabetic complications Drug use based on national drug Unclear how mortality was ascertained; MI and heart Cardiovascular disease, cerebrovascular disease, congestive heart failure, codes for prescription claims; mean failure data from Medicare claims previous insulin treatment, and nitrate use drug use around 215 days Use ascertained from pharmacy MI data from hospital discharge diagnosis and ICD Propensity score matching used with adjusted analysis for variety of claims database; mean follow-up code as primary event; "sudden death" events demographic and cardiovascular risk factors 8.4 months with regimen stop as captured through ambulance codes for resuscitation/ ACS=acute coronary syndrome; BMI=body mass index; CABG=coronary artery bypass graft; CAD=coronary artery disease; COPD=chronic obstructive pulmonary disease; eGFR=estimatedglomerular filtration rate; HbA1c=glycated haemoglobin; HDL=high density lipoprotein cholesterol; ICD=international classification of diseases; LDL=low density lipoprotein cholesterol;MI=myocardial infarction; NSAID=non-steroidal anti-inflammatory drug; PCI= percutaneous coronary intervention; TIA=transient ischaemic attack.
BMJ ONLINE FIRST bmj.com Study or subgroup retention, may explain its greater risk of congestive heart failure.39 40 Adjusted odds ratio 0.69 (0.28 to 1.69) Clinical and policy implications 1.14 (1.05 to 1.24) Our findings have important implications. Rosiglita- 1.16 (1.02 to 1.33) zone is still available on a restricted basis in the United Pantalone 200923 1.23 (0.79 to 1.92) States and Canada.8 41 However, for patients who need 1.36 (1.05 to 1.76) 1.02 (0.86 to 1.21).
thiazolidinedione treatment, continued use of rosiglita- Winkelmayer 200828 1.15 (1.05 to 1.26) zone may lead to excess heart attacks, heart failure, and Subtotal (95% CI) 1.14 (1.08 to 1.20) mortality, compared with pioglitazone. The size of the Test for heterogeneity: τ2=0.00, effect on public health may be considerable, given the χ2=4.83, df=6, P=0.57, I2=0% data from June 2009 showing that about 3.8 million Test for overall effect: z=5.05, P<0.001 annually in the United States.42 However, other Unadjusted odds ratio adverse effects are associated with both the thiazolidi- Lipscombe 200721 1.13 (0.92 to 1.38) nediones, such as the doubling of risk of fracture in Subtotal (95% CI) 1.13 (0.92 to 1.38) women.7 Concerns also exist about a modest increase Test for heterogeneity: Not applicable in the risk of bladder cancer with pioglitazone after Test for overall effect: z=1.15, P=0.25 long term use in an observational study and a higherpercentage of bladder cancers with pioglitazone rela- 1.14 (1.09 to 1.20) tive to comparator arms in long term randomised con- Test for heterogeneity: τ2=0.00, trolled trials.43 Further studies are needed to investigate χ2=4.85, df=7, P=0.68, I2=0% these other adverse events, as clinicians need to bal- Test for overall effect: z=5.18, P<0.001 ance these risks and benefits against those of emerging alternative agents such as incretin mimetics that may or may not be safer than thiazolidinediones.
Fig 4 Meta-analysis of odds ratio for overall mortality with rosiglitazone versus pioglitazone Limitations of study Our analysis has some limitations, relating mainly to As both drugs are known to cause heart failure, the the quality of the primary studies. Misclassification of increased risk of congestive heart failure associated with outcomes and drug use may occur in observational stu- rosiglitazone compared with pioglitazone represents its dies that rely on healthcare databases and discharge codes. However, any potential misclassification of for their differential effects on myocardial infarction are drug use and outcomes would affect both thiazolidine- possible. One possibility is that these findings represent diones equally. Non-randomised data are susceptible an ischaemic cardiovascular benefit with pioglitazone.
to selection bias and residual confounding. However, However, conclusive evidence on ischaemic cardio- investigators of the two largest cohort studies found vascular benefit with pioglitazone is lacking; a meta- little difference in the baseline demographics and analysis of trials yielded a relative risk of 0.81 (0.64 to cardiovascular risk of patients who used pioglitazone and rosiglitazone.17 19 Both drugs are from the same greater ischaemic cardiovascular hazard with rosiglita- class and were licensed for similar indications. Until zone, consistent with evidence from clinical trials.3 4 May 2007, no reasons existed why any specific groupof patients would have been systematically channelled Possible biological mechanism towards one thiazolidinedione or the other. Our risk The precise biological mechanisms responsible for estimates did not change despite exclusion of the two these differences in cardiovascular risk and mortalityare uncertain. Significant differences have been found between the thiazolidinediones in lipid metabolism;rosiglitazone causes greater elevations of triglycerides and low density lipoprotein cholesterol than does pioglitazone.36 Pioglitazone had a significantly more favourable effect on triglycerides, high density lipopro-tein cholesterol, low density lipoprotein particle con- centration, and low density lipoprotein particle size than did rosiglitazone. Whereas pioglitazone hasshown some potential benefit in preventing progres- sion of atherosclerosis,37 rosiglitazone failed to show any significant effect in preventing atherosclerosis in a recent study.38 The more powerful renal PPARγ ago- Fig 5 Funnel plot based on odds ratio for myocardial nistic effect of rosiglitazone, leading to more fluid infarction and study precision BMJ ONLINE FIRST bmj.com Lincoff AM, Wolski K, Nicholls SJ, Nissen SE. Pioglitazone and risk of WHAT IS ALREADY KNOWN ON THIS TOPIC cardiovascular events in patients with type 2 diabetes mellitus: ameta-analysis of randomized trials. JAMA 2007;298:1180-8.
Both rosiglitazone and pioglitazone are known to increase the risk of congestive heart failure Singh S, Loke YK, Furberg CD. Thiazolidinediones and heart failure: ateleo-analysis. Diabetes Care 2007;30:2148-53.
However, the risk of ischaemic cardiovascular events seems to have been mainly associated Loke YK, Singh S, Furberg CD. Long-term use of thiazolidinediones with rosiglitazone rather than pioglitazone and fractures in type 2 diabetes: a meta-analysis. CMAJ2009;180:32-9.
Any distinct differences between the cardiovascular effects of the thiazolidinediones have yet Cohen D. Insiders criticise FDA's decision not to withdraw to be fully clarified rosiglitazone. BMJ 2010;341:c5333.
Misbin RI. Lessons from the Avandia controversy: a new paradigm forthe development of drugs to treat type 2 diabetes. Diabetes Care WHAT THIS STUDY ADDS 10 Loke YK, Price D, Herxheimer A. Adverse effects. In: Higgins JPT, In observational studies, rosiglitazone is associated with increased odds of congestive heart Green S, eds. Cochrane handbook for systematic reviews of failure, myocardial infarction, and death compared with pioglitazone interventions. John Wiley & Sons, 2008.
11 Davies HT, Crombie IK, Tavakoli M. When can odds ratios mislead? studies that recruited after May 2007.17 19 Potential 12 Cates CJ. Dr Chris Cates EBM website. 2010. www.nntonline.net.
13 Deeks JJ, Higgins JP, Altman DG. Analysing data and undertaking exists for bias in selection of outcomes for analysis.
meta-analyses. In: Higgins JPT, Green S, eds. Cochrane handbook for However, although selective reporting favouring sig- systematic reviews of interventions. John Wiley & Sons, 2008.
14 Bilik D, McEwen LN, Brown MB, Selby JV, Karter AJ, Marrero DG,et al.
nificant beneficial outcomes may occur, one cannot Thiazolidinediones, cardiovascular disease and cardiovascular assume that reporting bias is similarly focused on sig- mortality: translating research into action for diabetes (TRIAD).
nificant findings of harm. The converse may occur with Pharmacoepidemiol Drug Saf 2010;19:715-21.
15 Brownstein JS, Murphy SN, Golfine AB, Grant RW, Sordo M, Gainer V, competing interests that emphasise interpretation and et al. Rapid identification of myocardial infarction risk associated reporting of safety in a manner that is favourable to with diabetes medications using electronic medical records.
Diabetes Care 2010;33:526-31.
rosiglitazone.44 Finally, we had insufficient data to 16 Dormuth CR, Maclure M, Carney G, Schneeweiss S, Bassett K, assess effects on stroke or death from cardiac causes.
Wright JM. Rosiglitazone and myocardial infarction in patientspreviously prescribed metformin. PLoS One 2009;4:e6080.
17 Graham DJ, Ouellet-Hellstrom R, MaCurdy TE, Ali F, Sholley C, Worral C, et al. Risk of acute myocardial infarction, stroke, heart Our results show that among patients with type 2 dia- failure and death in Medicare patients treated with rosiglitazone orpioglitazone. JAMA 2010;304:411-8.
betes, use of rosiglitazone is associated with a modest 18 Hsiao FY, Huang WF, Wen YW, Chen PF, Kuo KN, Tsai YW.
but statistically significant increase in the odds of myo- Thiazolidinediones and cardiovascular events in patients with type 2 cardial infarction, congestive heart failure, and death diabetes mellitus: a retrospective cohort study of over 473,000patients using the national health insurance database in Taiwan.
compared with patients receiving pioglitazone in real Drug Saf 2009;32:675-90.
world settings. Clinicians, patients, and regulatory 19 Juurlink DN, Gomes T, Lipscombe LL, Austin PC, Hux JE, Mamdani MM. Adverse cardiovascular events during treatment with authorities should carefully consider these results in pioglitazone and rosiglitazone: population based cohort study. BMJ the context of the available information on the thiazo- lidinediones' benefits on glycaemic control and harm 20 Koro CE, Fu Q, Stender M. An assessment of the effect of thiazolidinedione exposure on the risk of myocardial infarction in relating to different outcomes.
type 2 diabetic patients. Pharmacoepidemiol Drug Saf2008;17:989-96.
Contributors: YKL, CSK, and SS developed the concept and protocol for 21 Lipscombe LL, Gomes T, Levesque LE, Hux JE, Juurlink DN, Alter DA.
the review. CSK and YKL abstracted and analysed data. YKL, CSK, and SS Thiazolidinediones and cardiovascular outcomes in older patients wrote the manuscript. YKL had full access to all of the data in the study, with diabetes. JAMA 2007;298:2634-43.
takes responsibility for the integrity of the data and the accuracy of the 22 Margolis DJ, Hoffstand O, Strom BL. Association between serious data analysis, and is the guarantor.
ischemic cardiac outcomes and medications used to treat diabetes.
Funding: SS is supported by grant number 1KL2RR025006-03 from the Pharmacoepidemiol Drug Saf 2008;17:753-9.
23 Pantalone KM, Kattan MW, Yu C, Wells BJ, Arrigain S, Jain A, et al. The NCRR, a component of the National Institutes of Health (NIH), and NIH risk of developing coronary artery disease or congestive heart failure, Roadmap for Medical Research. The design and conduct of the study; and overall mortality, in type 2 diabetic patients receiving collection, management, analysis, and interpretation of the data; and rosiglitazone, pioglitazone, metformin, or sulfonylureas: a preparation, review, and approval of the manuscript was independent of retrospective analysis. Acta Diabetol 2009;46:145-54.
any sources of funding. Its contents are solely the responsibility of the 24 Stockl KM, Le L, Zhang S, Harada AS. Risk of acute myocardial authors and do not necessarily represent the official view of NCRR or NIH.
infarction in patients treated with thiazolidinediones or other Competing interests: All authors have completed the Unified Competing antidiabetic medications. Pharmacoepidemiol Drug Saf Interest form at http://www.icmje.org/coi_disclosure.pdf (available on 25 Tzoulaki I, Mokokhia M, Curcin V, Little MP, Millett CJ, Ng A, et al. Risk request from the corresponding author) and declare that (1) they have no of cardiovascular disease and all cause mortality among patients relationships with any company that might have an interest in the with type 2 diabetes prescribed oral antidiabetes drugs: submitted work in the previous three years; (2) their spouses, partners, retrospective cohort study using UK general practice research or children have no financial relationships that may be relevant to the database. BMJ 2009;3339:b4731.
submitted work; and (4) they have no non-financial interests that may be 26 Walker AM, Koro CE, Landon J. Coronary heart disease outcomes in relevant to the submitted work.
patients receiving antidiabetic agents in the PharMetrics database Data sharing: No additional data available.
2000-2007. Pharmacoepidemiol Drug Saf 2008;17:760-8.
27 Wertz DA, Chang CL, Sarawate CA, Willey VJ, Cziraky MJ, Bohn RL. Risk of cardiovascular events and all-cause mortality in patients treated Gale EA. Lessons from the glitazones: a story of drug development.
with thiazolidinediones in a managed-care populations. Circ Cardiovasc Qual Outcomes 2010;3:538-45.
Nissen SE, Wolski K, Topol EJ. Effect of muraglitazar on death and 28 Winkelmayer WC, Setoguchi S, Levin R, Solomon DH. Comparison of major adverse cardiovascular events in patients with type 2 diabetes cardiovascular outcomes in elderly patients with diabetes who mellitus. JAMA 2005;294:2581-6.
initiated rosiglitazone vs pioglitazone therapy. Arch Intern Med Nissen SE, Wolski K. Rosiglitazone revisited: an updated meta- analysis of risk for myocardial infarction and cardiovascular 29 Ziyadeh N, McAfee AT, Koro C, Landon J, Arnold Chan K. The mortality. Arch Intern Med 2010;170:1191-201.
thiazolidinediones rosiglitazone and pioglitazone and the risk of Singh S, Loke YK, Furberg CD. Long-term risk of cardiovascular events coronary heart disease: a retrospective cohort study using the US with rosiglitazone: a meta-analysis. JAMA 2007;298:1189-95.
health insurance database. Clin Ther 2009;31:2665-77.
BMJ ONLINE FIRST bmj.com 30 Lee CD, Folsom AR, Pankow JS, Brancati FL, for the Atherosclerosis rosiglitazone on progression of coronary atherosclerosis in patients Risk in Communities (ARIC) Study Investigators. Cardiovascular with type 2 diabetes mellitus and coronary artery disease: the events in diabetic and nondiabetic adults with or without history of assessment on the prevention of progression by rosiglitazone on myocardial infarction. Circulation 2004;109:855-60.
atherosclerosis in diabetes patients with cardiovascular history trial.
31 Nichols GA, Gullion CM, Koro CE, Ephross SA, Brown JB. The incidence of congestive heart failure in type 2 diabetes: an update.
39 Young PW, Buckle DR, Cantello BC, Chapman H, Clapham JC, Diabetes Care 2004;27:1879.
Coyle PJ, et al. Identification of high-affinity binding sites for the 32 Roussel R, Travert F, Pasquet B, Wilson PW, Smith SC Jr, Goto S, et al, insulin sensitizer rosiglitazone (BRL-49653) in rodent and human for the Reduction of Atherothrombosis for Continued Health (REACH) adipocytes using a radioiodinated ligand for peroxisomal Registry Investigators. Metformin use and mortality among patients proliferator-activated receptor gamma. J Pharmacol Exp Ther with diabetes and atherothrombosis. Arch Intern Med 40 Zhang H, Zhang A, Kohan DE, Nelson RD, Gonzalez FJ, Yang T.
33 Nissen SE, Wolski K. Effect of rosiglitazone on the risk of myocardial Collecting duct-specific deletion of peroxisome proliferator-activated infarction and death from cardiovascular causes. N Engl J Med receptor gamma blocks thiazolidinedione-induced fluid retention.
Proc Natl Acad Sci USA 2005;102:9406-11.
34 Song F, Altman DG, Glenny AM, Deeks JJ. Validity of indirect 41 Health Canada. Avandia, Avandamet and Avandaryl: important new comparison for estimating efficacy of competing interventions: restrictions on the use of rosiglitazone products due to information empirical evidence from published meta-analyses. BMJ on heart-related events. 2010. www.hc-sc.gc.ca/dhp-mps/medeff/ 35 Papanikolaou PN, Christidi GD, Ioannidis JPA. Comparison of evidence on harms of medical interventions in randomized and 42 Shah ND, Montori VM, Krumholz HM, Tu K, Alexander GC, nonrandomized studies. CMAJ 2006;174:635-41.
Jackevicius CA. Responding to an FDA warning—geographic variation 36 Goldberg RB, Kendall DM, Deeg MA, Buse JB, Zagar AJ, Pinaire JA, et in the use of rosiglitazone. N Engl J Med 2010;10:1056.
al, for the GLAI Study Investigators. A comparison of lipid and 43 US Food and Drug Administration. FDA drug safety communication: glycemic effects of pioglitazone and rosiglitazone in patients with ongoing safety review of Actos (pioglitazone) and potential increased type 2 diabetes and dyslipidemia. Diabetes Care 2005;28:1547-54.
risk of bladder cancer after two years exposure. 2010. www.fda.gov/ 37 Nissen SE, Nicholls SJ, Wolski K, Nesto R, Kupfer S, Perez A, et al, for the PERISCOPE Investigators. Comparison of pioglitazone vs 44 Wang AT, McCoy CP, Murad MH, Montori VM. Association between glimepiride on progression of coronary atherosclerosis in patients industry affiliation and position on cardiovascular risk with with type 2 diabetes: the PERISCOPE randomized controlled trial.
rosiglitazone: cross sectional systematic review. BMJ 38 Gerstein HC, Ratner RE, Cannon CP, Serruys PW, GarcÃa-GarcÃa HM, van Es GA, et al, for the APPROACH Study Group. Effect of Accepted: 15 February 2011 BMJ ONLINE FIRST bmj.com
Primary Care 1st Edition Use of Antipsychotics in Behavioural and Psychological Symptoms of Dementia (BPSD) This tool is designed to help providers understand, assess, and manage patients in primary care with behavioural and psychological symptoms of dementia (responsive behaviours), with a focus on antipsychotic medications. This tool integrates best-practice evidence with clinical experience, and makes reference to relevant existing tools and services wherever possible.