Kazim Sheikh

Effect of Measurement on Sex Difference in Stroke Mortality

The 1994 to 1997 administrative data on 40 450 elderly Medicare beneficiaries and general population of 2 states were used to measure “case mortality” (deaths attributable to any cause among cases of acute stroke), “case fatality” (deaths caused by cerebrovascular diseases among cases of acute stroke), and “population mortality” (deaths caused by stroke in the elderly general population). Mortality was higher in men than in women according to all measures except population mortality caused by subarachnoid hemorrhage. There was no sex difference in 1-year case fatality. One-year all-cause mortality among cases of nonhemorrhagic stroke or all types of stroke was higher in men than in women. Similar sex differences were found in 4-year population mortality caused by nonhemorrhagic stroke or all types of stroke combined. The 3 measures differed with respect to sex difference in stroke mortality. How stroke is defined and how mortality is measured does affect sex difference.

Rise and fall of radical prostatectomy rates from 1989 to 1996.

OBJECTIVES To describe the changes in the rates of radical prostatectomy procedures, prostate-specific antigen (PSA) screening tests among Medicare beneficiaries, and the incidence of prostate cancer in the United States and to explain the exaggerated increase and decrease in the frequency of radical prostatectomy from 1989 to 1996. METHODS Medicare claims data on radical prostatectomy procedures and screening PSA tests and the National Cancer Institutes Surveillance, Epidemiology, and End Results prostate cancer incidence data were used to estimate the rates of PSA testing and radical prostatectomy among Medicare beneficiaries aged 65 to 74 years and 75 years and older (population rates). The age-specific true rates of the procedure were also estimated for the incident cases of prostate cancer (the population at risk of undergoing radical prostatectomy) among the beneficiaries. RESULTS The PSA test, prostate cancer incidence, and radical prostatectomy rates increased from 1989 to 1992. Thereafter, the incidence of prostate cancer, and the population and true rates of radical prostatectomy declined. The percentage of increase and decrease in the population rate of radical prostatectomy was approximately twice that in its true rate. CONCLUSIONS The radical prostatectomy rates based on all Medicare beneficiaries grossly exaggerated the changes in the use of the procedure. Where possible, true rates, using the population at risk as the denominator, should be used in the studies of diagnostic and therapeutic procedures, complications, and adverse effects.

Variation and changes in state-specific carotid endarterectomy and 30-day mortality rates, United States, 1991-2000

OBJECTIVES The objectives of this study were to investigate variations between states and changes in state-specific carotid endarterectomy (CEA) and 30-day mortality rates. Cross-sectional variations and changes over time in such measures may be indicative of improvement in the quality of care. METHODS We performed retrospective analyses of pre-existing administrative data on Medicare beneficiaries aged 65 years and older in the United States. Age-adjusted, state-specific CEA rates and 30-day postoperative mortality rates in 1991, 1995 and 2000 were examined, as well as changes in these rates from 1991 to 1995 and from 1995 to 2000. Stroke mortality in the general population of each state was used as a crude measure of the need for CEA procedure in the state. The Spearman rank correlation analysis was used to study correlations between rates. Oldhams method was used to avoid the effect of regression to the mean. RESULTS There were wide variations in the state-specific CEA rates, 30-day mortality, and in changes in these rates over time. The states with relatively low procedure rates in 1991 also had low rates in 1995 and 2000, and relatively higher increases in the rates. The states with relatively high 30-day mortality in 1991 or 1995 had lower increases or greater decreases in the rate. CEA rates were not correlated with any measure of surgical mortality, but they were correlated with stroke mortality in the general population. CONCLUSIONS The inter-state variation in CEA rates has not changed much since 1991, but variation in 30-day mortality decreased through 2000. The states with low procedure rates in 1991 did not have sufficient increase to catch up with the high-rate states by 1995, but they were prone to experience a higher increase in the subsequent 5 years. The validity of stroke mortality in a state as a measure of the need for CEA is questionable. Further research using clinical data is needed to better explain variations between states.

Utility of provider volume as an indicator of medical care quality and for policy decisions

Associations between hospital volume or physician caseload and patient outcome have been used to assess the performance of health care providers. Although most studies have focused on major surgical procedures, in-hospital or 30-day mortality from many nonsurgical conditions and procedures has also been examined. Although high volume may be a surrogate for the providers skill and experience, and better outcomes may attract greater volumes, aggregate data on provider volume show many outliers indicating that the outcome for some low-volume providers is better than that for high-volume providers. Mortality is only one measure of medical care quality. Although high volume may not always be indicative of favorable outcome, referral of patients from low-volume to high-volume providers has been recommended. It has also been suggested that patients choose health care providers on the basis of physician caseload. It is unclear how such recommendations could be implemented in practice; furthermore, they would deprive many patients from access to, as well as disrupt the provision of, adequate health care in many areas. An alternative to requiring patients to receive care from high-volume providers is to adopt other measures for improving outcomes, such as improving the quality of care provided by low-volume providers and attracting better providers to low-volume areas.

Total Cholesterol, Severity of Stroke, and All-Cause Mortality

To the Editor: The association of blood cholesterol with the risk of stroke, a very important clinical and public health issue, appears to be in dispute. To fuel the debate, Olsen et al1 reported a study of 513 patients admitted in a Copenhagen, Denmark hospital with acute ischemic stroke. All study data, including measures of total serum cholesterol (TSC) and the severity of stroke, were collected on admission in the hospital, and poststroke 10-year all-cause mortality was ascertained. The severity of stroke was measured by the Scandinavian Stroke Scale that is based on a composite of neurological deficits, and severity was found to be correlated with the cerebral infarct size seen on CT scans. The authors found that poststroke TSC levels were inversely associated with the severity of stroke, cerebral infarct size, and mortality, which were adjusted for the covariates.1 The authors concluded that “hypercholesterolemia primarily is associated with minor strokes due …

Investigation of selection bias using inverse probability weighting

In 1999-2001, university graduates in the city of Pamplo na, Spain were invited to participate in a study of cardio vascular diseases and motor vehicle injuries [1]. This selection prevented generalization of the results to the population of Pamplona. By January 2002, 9907 self-se lected graduates (the study cohort) from an unknown sampling frame returned mail questionnaires containing the baseline data. Two years later, 13% of the cohort failed to respond to the follow-up questionnaire. For the purpose of the study of hypertension, 2767 subjects were excluded from the analysis because of their prevalent hypertension and selected comorbid conditions, outlying values of caloric intake, or missing values for study variables in the baseline and/or follow-up data [1], yet another selection. Alonso et al. [1] chose to address selection bias due to non-response to follow-up. They studied the association between body mass index (BMI) in non-hypertensive subjects and subsequent hypertension adjusted for con founding and non-response by inverse probability weight ing (IPW). Non-respondents were different from respondents with respect to missing values in the baseline data and measures of BMI and covariates in addition to the

Statins and risk of incident diabetes

The meta-analysis of randomised controlled trials is the apotheosis of scientifi c enquiry in clinical medicine. Naveed Sattar and colleagues (Feb 27, p 735) diligently apply this method to the vexing question of whether statin therapy aff ects incident diabetes. They report that statins increase the odds of incident diabetes. This fi nding is contrary to that of an original report by many of the same authors in 2001. With non-adherence rates of up to 20% among the approximately 24 million patients receiving statin therapy in the USA, it behoves us to question these results rigorously before their dissemination to the popular media. Reported limitations of this metaanalysis include the loss of statistical signifi cance on exclusion of the two trials in which incident diabetes was by physician report only (CORONA and HPS). Sattar and colleagues should also have excluded the JUPITER data here (the numbers used [270 vs 216] were actually physician reported). Non-reported limitations include the use of post-hoc analysis, with only three of the 13 trials including incident diabetes as a prespecifi ed endpoint (HPS, ASCOT, LIPID). Subgroup analysis was done in 12 trials (only JUPITER excluded patients with diabetes). Post-hoc and subgroup analyses introduce analytical errors that can lead to misleading conclusions. Further, quantitative interactions in which treatment eff ects are consistent are more credible than are interactions in which treatment eff ects vary between trial subgroups (see fi gure 3 of Sattar and colleagues’ meta-analysis). Therefore, these capricious fi ndings should be regarded as a hypothesisgenerating statistical anomaly, lest we fall foul of the logical fallacy, “Post hoc ergo propter hoc”. I declare that I have no confl icts of interest.

Effectiveness of Interventions for Reducing the Frequency of Radical Prostatectomy Procedures in the Elderly: An Evaluation

Between 1993 and 1997, the Peer Review Organizations (PROs) implemented interventions for reducing radical prostatectomy rates in 50 selected hospitals in 10 states and all hospitals in an additional 4 states. Control hospitals and states were matched with the intervention hospitals and states. Prostate cancer incidence rates were used to estimate the number of Medicare beneficiaries aged 75 years and older with prostate cancer, the denominators for the procedure rates, in the hospital service area of each intervention and control hospital, and in each state and their controls. After interventions, significant reductions in the state-specific radical prostatectomy rates were achieved in the intervention hospitals in 2 states and in 1 of the 4 intervention states where statewide interventions had been implemented. Similar reductions were seen in the control hospitals in 3 other individual states and 8 states combined where hospital-based interventions were implemented. These changes in the procedure rates were most likely due to the national decline in the incidence of prostate cancer, not the PRO interventions.

Metabolic Syndrome and Stroke

To the Editor: In an interesting study of 991 members of a Finnish cohort,1 Wang et al found that 5 definitions of metabolic syndrome (MS) were associated with the risk of all types of stroke combined, and 3 components of MS—glucose intolerance, insulin resistance and central obesity—were independently associated with stroke. The sixth definition (American College of Endocrinology [ACE] definition) of MS did not include obesity as a component of MS but raised body mass index was regarded as a risk factor for MS.2 If Wang et al had restricted their study to ischemic stroke, they might have found MS by ACE definition also to be associated with stroke. Their findings are not surprising because the components of MS (obesity, hypertension, …

Prevalence of Intracranial Atherosclerosis and Coexistent Conditions in Ethnic South Asians

To the Editor: Although the objectives of their study were not stated, De Silva et al1 used clinical data on 200 acute ischemic stroke patients admitted in a hospital in Singapore to determine the prevalence of intracranial large-artery disease (ICLAD) and its risk factors. The study subjects were selected on the basis of their origin of the Indian subcontinent (ethnic South Asians). The authors declared that “ICLAD is the most common vascular lesion in stroke worldwide”. To justify their declaration, they cited a reference (Reference 2 in their article1) to an article published in Int J Stroke . This publication was not found in PubMed Central. The authors found that 54% of stroke patients had ICLAD and they pronounced it as “high burden of …