复旦大学：《循证医学》课程教学资源（预习资料）疾病预后证据的分析评价预习.pdf_大学文库

The Medical Literature Users' Guides to the Medical Literature V. How to Use an Article About Prognosis Andreas Laupacis, MD, MSc: George Wells, MSc, PhD; W. Scott Richardson, MD: Peter Tugwell, MD, MSc for the Evidence-Based Medicine Working Group CLINICAL SCENARIO the dementia. You indicated that you nying the disease in question). They can pa You are about to see a 76-year-old would discuss this with him at the sec- predict any outcome, whether good (eg, retired schoolteacher for the second ond visit once the results of all the tests cure or survival) or bad (eg, death or To stud time. You first saw her in the clinic a are available complication). Prognostic factors need month ago because of cognitive prob- seArCh not necessarily cause the outcomes, just iduals wh eluded a Standardized Mini-Mental State Hoping to provide the son with the to predict their development. In the li- ontrol Examination,'on which she scored 18 his mother's prognosis, after the initial distinguished from"risk factors,"those thenumbe out of a possible 30 points, and a physi al examination that was normalinclud-. visit you searched the medical library,s patient characteristies associated with i ith a par ng no focal neurological signs. You ar- MEDLINE CD-ROM system via the. the development of the disease in the ranged investigations for the treatable: hospitals network on the clinic com- first place. For example, smoking is an more and you thus feel she has probable alz- ease "which yielded 3687 articles from ment of lung cancer, but is not as im he poter portant a prognostic factor as tumo The patient has lived with her son nosis, "which yielded 28004 articles;; stage in someone who has lung cancer /ses and heimer disease. nificant problems with her memory found several articles of potential inter- tic factors. Therefore, the best study de: relatives c about 3 years ago. However, she has est, including one that seemed precisely sign to identify the presence of and de quarts), im noid during the last year. She has re- Alzheimer-Type Dementia "by Walsh a prognostic factor is a cohort study. As sadie A fused to allow him to look after her fi..and colleagues. 2 nancial affairs, despite the fact that she series, investigators conducting cohort shut owns three pieces of property and isn't INTRODUCTION study follow one or more groups( cohorts able to manage them herself. Her son In this article we will suggest a frame- of individuals who have not yet suffered ase-cont hether she is likely to die soon' from sess articles that deal with prognosis, berofoutcome events over time. Anidea b prate whe using the article on patients with de- cohort study consists of a well-defined equired dr mentia as an example. We will follow sample of individuals representative Department of Clinical Fni Axiology the usual format of this series and dis- the population of interest and uses oP-.ARE THE F cuss how to determine whether the re- jective outcome criteria (Dr Laupacis): the Departments of Medicine and Epi- sults are valid, how to interpret the re- conducted in Framingham, Mass, in which THE ST semiology and Community Medicine, University of Ot- sults, and whether the information will investigators have followed a cohort ry G 5209individuals since 1948, has provided Was The (NY)School of Medicine and aprognosis'refers to the possible out. clinicians with a great deal of useful ir I Nell-Defir A complete list of members(with afflictions) of the comes of a disease and the frequency formation about the prognostic impor similar Po the first article of this series (AMAroup 800293. with which they can be expected to oc- tance of many determinants of can article: Gordon H. Guyat ( chair MD. MSc: George tia). Sometimes the characteristics of a ized trials include careful documentation /ed. The Gerstein, MD MSc: Brian. Haynes, MD. MSc, PhD: particular patient can be used to more of inclusion criteria and strict protocols and whe ir 3. MPH: Mitchell Levine, MD). MSc accurately predict that patient's even- for follow-up, patients in such trials form pheunderly Jim Nishikawa, MD: David. Sacket, MD MS: Patrick tual outcome(eg, a patient with demen- cohorts that can also generate informs mould desc HUrLe . sc: stephane sauve, MD, sc: virginia worse prognosis than someone without However, the patients entered into the David Naylor, D, Dp il: Andrew behavioral problems). These character- trial are often not representative of the Azh disorde elected th∈ eral b types, such as demographic(eg, age), morbid (eg, other conditions accompa-.JAMA 234JAMA,J20,1994vo272.No.3 Users' Guides to Medical Literature-Laupacis e

Guides for an Articie About Prognosis from primary to tertiary centers raises long period, investigators must follow re the results of the study valid usual cases, thus increasing the likeli. patients for long enough to detect the Was there a representative and well-defined hood of adverse or nonfavorable out- currence in some women with early Was follow-up sufficiently long and complete? filter bias, "the likelihood of a subse- ter initial diagnosis and treatment. Pa- quent nonfebrile seizure in children with tients in the dementia study were en- their first febrile seizure was much lower rolled between 1980 and 1982 and fol- Was there adjustment for important prognostic in community-based populations than in lowed until 1988 or their death. Thus What are the results? those drawn from hospitals. the follow-up was quite long, and 61%of The second issue concerns whether the cohort died during this time the study patients are all at a similar, Ideally, investigators will succeed in the results help me in caring for my pa- well-defined point in the course of their following all patients(as they did in the disease. Authors should provide a clear. dementia study) but this is often not the to my own? the results lead directly to selecting or avoid- description of the stage of disease at case. Patients are not usually unavail- Are the results useful for reassuring or counseling instance, since the duration of illness is reasons. Patients may fail to return be- ey can often associated with outcome, the in- cause they have suffered exactly those vestigators should report the duration. events in which the investigators are ath or To study prognostic factors, investi- ofillness for the sample patients Within interested (eg, they died or have been ators can also collect"cases"of indi- reason, all or most of the study patients institutionalized). Conversely, patients es, just iduals who have already suffered the should be at a similar point, such as who feel entirely healthy may also be enough outcome event and compare them with survivors of a first myocardial infarc-. less likely to return for evaluation be- the lit ntrols” who have not. In these“case usually ontrol"studies the investigators count lung cancer. However, the similar point greater the number of patients unavail- those the number of individuals in each group:on. in the course of disease need not be early able for follow-up, the less accurate the 计 I a particular prognostic factor(eg estimate regarding the risk of the ad in the were the patients with dementia who Walsh and colleagues2studied 126- verse outcome. ig is an, died more likely to have had behavioral patients with Alzheimers disease who Under what circumstances does un- °如 the and controls, as well as the ret- twee9 diagnosis was you consider the relation between the velop- I problems than those who did not die?). were consecutively referred to a mul- availability for follow-up compromise the thia to prognostic factors (which often depends ing of an internist, psychiatrist, psycholog able and the proportion of patients wh orognos. on the memory of the patients or their gist, neurologist or neuropathologist, and have suffered the adverse outcome of tudy de- relatives or the accuracy of medical research nurse using the conventional interest. The larger the number of pa- and de. charts), limits the strength of inference Diagnostic and Statistical Manual of tients whose fate is unknown to ed with clinicians can draw from case-control. Mental Disorders, Fourth Revision cri- the number who have suffered an event teria for, dementia. The tests used to the greater the threat to the studys cudy, As I studies. Also, case-control studies can- exclude other causes of dementia were validity. Forinstance, letus assume that not provide information about the ab- a cohort solute risk of an event, but only about not described. However, given the mul- 30% of a particularly high-risk group (cohorts) I the relative risk (RR). Nevertheless, tidisciplinary nature and expertise of. (such as elderly diabetics)have suffered se-control studies can provide useful. the group, it seems reasonable to as-. an adverse outcome (such as cardiovas- Anidea f priate when the outcome is rare or the done to exclude disorders such as hy-. up. If 5% of the patients have been lost, defined i required duration of follow-up is long. pothyroidism, depression and space-oe. the true rate of patients whe khad died uses RE THE RESULTS fused with Alzheimers diseas so, the clinical implications would not ort stud T OF THE STUDY VALID? Walsh and colleagues reported sur- change, and the unavailability for follow vival from two different points in time: up doesn't threaten the validity of the (1)referral to the clinic and(2)the point study. However, in a much lower-risk provided Was There a Representative and at which symptoms of memory loss were patient sample(otherwisehealthy useful in Well-Defined Sample of Patients at a first noticed. The former is a more cer- middle-aged men, for instance)the ob- ic impor-I imilar Point in the Course of the Dis- tain point in time, but suffers from the served event rate may be 1%. In this of cardio ease?-This guide addresses two related disadvantage that patients cometo medi- case, if one assumed that all 5% of the mentation ined the individuals in the study are, progression of their disease. The latter died, the event rate of 6% would have protocols and whether they are representative of provides a more uniform starting point, very different implications. If the num rials form I the underlying population. The authors but is potentially imprecise because de- ber of patients unavailable for follow-up e informs should describe and specify their crite- entia develops insidiously and the time potentially jeopardizes the study,'s va establishing that the patient has of onset is identified retrospectively. lidity, you should look for the rea ad into the i the disorder of interest (in this case, Survival after presentation to clinic is for patients being unavailable, and com- tive of th azheimer-type dementia) and how they probably more relevant for your pa- pare the important demographic and lected their patient sample cteristic of the patic Was Follow-up Sufficiently Long who were unavailable with the patients by of the patients can distort the re- and Complete -since the presence of whom follow-up was complete. To the e Sections: a)ts of a study. For example the se-. a prognostic factor often precedes the extent that the reasons for disappear- ice of referrals that leads patients development of an outcome event by a ance are unrelated to outcome and the aupacs'eta AMA, Juth201940272.No.3 Users'Guides to Medical Literature- - Laupacis et al 235

wuwsucu.Llllives-tients into a number of cohorts del tigators omit information about reasons ing on their demographic characteris- chme宽ym的 w-up or the tics, disease variables,and!8 for unavailab Dle, the strength of in t sor graphic variables from the study results will be weaker. level) and functional status (eg, modi- Streptokinase nd Aspirin Secondary Guides fied walking time and activities of daily Were Objective and Unbiased Out. talit living) were strongly predictive of mor- come Criteria Used? -Investigators Since treatments can also alter pa- must provide a clear and sensible defi-. tient outcomes, they should be taken Placebo Indus nition of adverse outcomes before the into account when analyzing prognostic study starts Outcome events can vary factors. While such treatments are not Time From Randomization, mo.sam measured(death), to those that recom, ences in treatment in the analysis For bents teat d with streptokinase ann action h -a h ment and may often be difficult to mea- that examined the prognosis of Q-wave Group. Randomised trial of intravenous strepto International Study of Infarct Survival)Collaborativ aury minimize bias, the individual determin-tion, the investigators adjusted for age, cases of suspected acute myocardial infarcion whetherthe apotential prog. angina pectoris, congestive heart fail nostic factor. This is not always possible ure, and cardiovascular disease prior to and, for unequivocal events such as the infarct. However, they did not take death, may not be necessarv However, into account. treatment with aspirin or blinding is essential for outcomes re-B-blockers, which clinicians may have quiring a great deal of judgment, such administered at the time of the infarct as transient ischemic attacks or unstable and which we know have an impact on a50 angina. In the study by Walsh and col- mortality pth1mm矿ten/ he study was known for those whodied mentia. However, they did evaluate the Was There Adjustment for Impor- importance of 20 potential prognostic tant Prognostic Factors?When com- factors in their cohort. Age at symptom. Fig2.Need for revision after total hip arthrpist sive oft patients, investigators should consider falling, behavioral problems, and hear whether their clinical characteristics are ing loss all had a statistically significant (adapted from Dorey and Amstutz). similar and adjust the analysis for any relation to mortality differences they find. The Framingham discuss some of the prognostic factis Cl for Study investigators reported that the WHAT ARE THE RESULTS? rate of stroke in patients with atrial fi- disease. As mentioned previously, t rpolat The quantitative results from studies statistically significant prognostic fa A ote th brillation and rheumatic heart disease : of prognosis or risk are the number of tors for death were increasing age c was 4l per 1000 person years, which events that occur over time. We will mentia severity, behavioral problem. merenre was very similar to the rate for patients describe three common expressions of wandering and falling, and hearing tut'rty o sta heart disease However, patients with mentary information about prognosis. erably older than his uncle was at th[ momper than those who did not have rheumatic Outcome Event(s) in a Specified Pe- explains some of the difference. It won I petients w heart disease. Once adjustments were riod ofTime?--Yourpatient's sonasked, be nice to use the prognostic factor i a Med.I. made for the age, sex, and hypertensive."What are the chances that my mother: further refine the chance of death in bi/a lern status of the patients, the investigators will still be alive in 5 years? You can mother. Her age is almost identica I aides to ec greaterin patients withrheumatic heart absolute terms. Five years after pre- Walsh and colleagues. Howeve disease and atrial fibrillation than in pa- sentation to the clinic about one-half the Mini-Mental State Examination scores m b p iBE tients with atrial fibrillation who did not patients(50%)had died. Thus, there is quite low (indicating more severe eI p bennett Ju have rheumatic heart disease about a 50: 50 chance that his mother mentia), and her behavioral proble any studies of prognosis break the will be alive in 5 vea pected prognostic factors. Comparison cate that the only person he knows with Walsh et al. Unfortunately, no tabie the natur of the pattern and frequencies of out- Alzheimer's disease is a 65-year-old uncle formula was presented that allot comes between these grouth the prog." living Heis surprised that his moth- timate a risk of mortality that is spear mine the RR associated an deter- who was diagnosed 10 years ago and is to combine all of these factors and ow seatietiad nostic factor in question. For example, er's chance of dying in the next 5 years for your patient. However, you can l i Pincus and colleagues followed a co- is so high. This gives you the chance to. confidentin telling her son that his 236 JAMA, July20,194Vo272,№.3 care'Gi ides to Medical Literature at

chances of dying are at least 50% earlier follow-up periods usually include tic data often provide the basis for sen- during the next 5 years, and probably results from more patients than the later sible decisions about therapy. Knowing periods(because ofunavailability for fol- the expected clinical course of your pa- ondition can help yo 出aeor, urvival could rolled into the study址 the same time9) whether treatment should be offered a址 r relative expressions of results ad- more precise in the earlier periods, in-. creases the risk of stroke in patients gs this question. For this answer we dicated by narrower confidence bands with nonrheumatic atrial fibrillation and should turn to a survival curve, a graph around the left-hand parts of the curve. is indicated for many patients with this he number of events over time(or Walsh and colleagues also provided disorder b However, in one study the I onversely, the chance of being free of 95%0 CIs for the reassociated with each frequency of stroke in patients with t be discrete(eg, death, stroke, re-. associated with a behavioral problem years of age or younger with no asso- laurence of cancer), and the time at was 1.5 with a 95%o CI of 1.0 to 2.5. This ciated cardiopulmonary disorders) was hich they occur must be precisely means that the best estimate is that a. 1.3%. over 15 years. 6 The risks of long Known. In most clinical situations the patient with a behavioral problem is 1.5 term warfarin therapy in this group of n in pe.nance of an outcome changes withtime. times more likely to die than an indi-. patients probably outweigh the'benefits s 1 and 2 show two survival vidual without a behavioral pi Are the Results Useful for Reassur- Second lairves, one of survival after a myocar- probability that the true RR is between ing or Counseling Patients?-Even if epok. ial infarction and the other the re- 1.0 (ie, no effect)and 2.5 is 95% the prognostic result does not lead you 139ear that the chance of dying after a myo- WILL THE RESULTS HELP ME IN to prescribe an effective therapy, it can still be clinically useful. A valid, precise and generalizable result of uniformly ter the event (reflected by an initially Were the Study Patients Similar to good prognosis is very helpful to the ep slope of the curve, which then flat- My Own?-How well do the study re- clinician when reassuring a concerned lens), while very few hip replacements sults generalize to the patients in your: patient orrelative. Some conditions, such quire revision until much later(this practice? The authors should describe as asymptomatic hiatal hernia or asymp- Walsh and colleagues provided a sur- allow comparison with your patients. The. good overall prognosis that they have al curve in Fig 1 of their article that article should list the patients'impor- been termed"nondisease. Onthe other suggests that the chance of dying is more tant clinical characteristics, along with. hand, a prognostic result of uniformly lor less constant during the first 7 years the, definitions used for these charac- bad prognosis provides the clinician with Precise Are the Estimates of the patient before you and those in the patient and family, leading to counsel- 9 Likelihood?-Even when valid, a prog- study, the more confident you can be in ing about end-of-life concerns In your patient, information on the s of the true risk. After determining the tient. The characteristics of the study likelihood of death will be useful to the easy sae of the risk, we should next examine patients were quite similar to your pa- son and his family as they plan the fu- Will the Results Lead Directly to: prognostic information about the rate of Walsh and colleagues found that the 95%. Selecting or Avoiding Therapy?-Since. progression of the dementing process factors CI for survival 5 years after presenta- there are no therapies for dementia that and the need for intensive, nursing care sly, the -apolated from Fig 1 in their article). fective, this guide does not directly ap- we thank Malcolm Hing, MiD, for his comments ng loss oloy DW, Alemayehu E, Roberts R 过 a standardize ompared with the traditional Mini-Mental State Group. Systemic treatment of early breast cancer intravenous streptokinase, oral aspirin, both,or 9:1-15. th in is lesie M walter S, Lee E, et al, for the Evi, B, Wol, Dawber TR, Thomas HE, Kannel WB Joint Surg Am. 1989; 71: 544-545 died b: aricle about harm. JAMA. 1994( 271:1615-1619 lation and risk of stroke: the Framingham Study. Antithrombotic therapy in atrial fibrillation. Chest, 992102:4265-4 Awber TR, Kauinel WB, Lyell LP. An approach 10. Pincus T, Brooks RH, Callahan LF. Prediction 16. Kopecky SL. The natural history of lone atrial 4556.arthritis according to simple questionnaire adjoint decades. N Engl J Med. 1987: 817: 669-6.three 17. Meador CK. The art and science of nondisease roblems I Fiey of the institute of Medicine. Inclusion of. 11. Berger C, Murabito JM, Evans JC, Anderson N Med.1965:27292 8. Stern Y, Mayeux R, Hauser WA, Bush T, Pre table ag PeN AMA 1d of disease: studies of febrile JAMA.1992;2281545-1551 Drachman DA, O'Donnel bf, Lew rA, swearer o I Statistical Manual of Mental Disorder. 4th: a t) regression. J Gen Intern Med 1993: 8 12. Katz M, Hauck ww. Proportional JM. The prognosis in Alzheimers disease: how far' ather than how fast' best predicts the course. Arch Neurol.1990;47:51-856 84Jy20,194Vol272No3 Users' Guides to Medical Literature--Laupacis et al 237

The NEW ENGLAND JOURNAL Of MEDICINE ORIGINAL ARTICLE Obstructive Sleep Apnea as a risk factor for Stroke and death H. Klar Yaggi, M.D., M.P. H, John Concato, M.D., M.P.H Walter N. Kernan, M D, Judith H. Lichtman, Ph. D, M.P.H Lawrence M. Brass, M. D. and Vahid Mohsenin. M.D BACKGROUND From the Section of Pulmonary and Crit. Previous studies have suggested that the obstructive sleep apnea syndrome may be an ae icine Me.kin. M.,. the section ef important risk factor for stroke It has not been determined, however, whether the syn to Departments of Epidemiology and Public ment for other risk factors, including hypertension. (LM. B ) Yale University School of Medi- METHODS of Pulmonary and Critical Care Medicine In this observational cohort study, consecutive patients underwent polysomnography, Sian of neurology (LMB). eterans ef. structive sleep apnea syndrome was based on an apnea-hypopnea index of 5 or higher airs Connecticut Healthcare System, West (five or more events per hour); patients with an apnea-hypopnea index of less than aven, Conn Address reprint requests to 5 served as the comparison group. Proportional-hazards analysis was used to determine Me icons non t the ta e enter po ep the independent effect of the obstructive sleep apnea syndrome on the composite out- 208057, New Haven, CT 065 come of stroke or death from any cause N Engl J Med 2005: 353: 203 Society RESULTS Among 1022 enrolled patients, 697(68 percent) had the obstructive sleep apnea syn drome. At baseline, the mean apnea-hypopnea index in the patients with the was 35, as compared with a mean apnea-hypopnea index of 2 in the comparison group In an unadjusted analysis, the obstructive sleep apnea syndrome was associated with stroke or death from any cause(hazard ratio, 2. 24: 95 percent confidence interval, 1.30 to 3.86; P=0.004). After adjustment for age, sex, race, smoking status, alcohol-consump- tion status, body-mass index, and the presence or absence of diabetes mellitus, hyper lipidemia, atrial fibrillation, and hypertension, the obstructive sleep apnea syndrome retained a statistically significant association with stroke or death(hazard ratio, 1.97; 95 percent confidence interval, 1. 12to 3.48; P=0.01).Ina trend analysis, increased sever ity of sleep apnea at baseline was associated with an increased risk of the developmen of the CONCLUSIONS death fro tension 第85页 Downloadedfromwww.nejm.orgonJanuary23,2010.Copyright@2005MassachusettsMedicalSocietyAllrightsreserved

original article The new england journal of medicine 2034 n engl j med 353;19 www.nejm.org november 10, 2005 Obstructive Sleep Apnea as a Risk Factor for Stroke and Death H. Klar Yaggi, M.D., M.P.H., John Concato, M.D., M.P.H., Walter N. Kernan, M.D., Judith H. Lichtman, Ph.D., M.P.H., Lawrence M. Brass, M.D., and Vahid Mohsenin, M.D. From the Section of Pulmonary and Critical Care Medicine, Yale Center for Sleep Medicine (H.K.Y., V.M.), the Section of General Medicine (J.C., W.N.K.), and the Departments of Epidemiology and Public Health (J.H.L., L.M.B.) and Neurology (L.M.B.), Yale University School of Medicine, New Haven, Conn.; and the Section of Pulmonary and Critical Care Medicine (H.K.Y.), the Clinical Epidemiology Research Center (H.K.Y., J.C.), and the Section of Neurology (L.M.B.), Veterans Affairs Connecticut Healthcare System, West Haven, Conn. Address reprint requests to Dr. Mohsenin at the Yale Center for Sleep Medicine, 300 Cedar St., TAC 441, P.O. Box 208057, New Haven, CT 06520. N Engl J Med 2005;353:2034-41. Copyright © 2005 Massachusetts Medical Society. background Previous studies have suggested that the obstructive sleep apnea syndrome may be an important risk factor for stroke. It has not been determined, however, whether the syndrome is independently related to the risk of stroke or death from any cause after adjustment for other risk factors, including hypertension. methods In this observational cohort study, consecutive patients underwent polysomnography, and subsequent events (strokes and deaths) were verified. The diagnosis of the obstructive sleep apnea syndrome was based on an apnea–hypopnea index of 5 or higher (five or more events per hour); patients with an apnea–hypopnea index of less than 5 served as the comparison group. Proportional-hazards analysis was used to determine the independent effect of the obstructive sleep apnea syndrome on the composite outcome of stroke or death from any cause. results Among 1022 enrolled patients, 697 (68 percent) had the obstructive sleep apnea syndrome. At baseline, the mean apnea–hypopnea index in the patients with the syndrome was 35, as compared with a mean apnea–hypopnea index of 2 in the comparison group. In an unadjusted analysis, the obstructive sleep apnea syndrome was associated with stroke or death from any cause (hazard ratio, 2.24; 95 percent confidence interval, 1.30 to 3.86; P=0.004). After adjustment for age, sex, race, smoking status, alcohol-consumption status, body-mass index, and the presence or absence of diabetes mellitus, hyperlipidemia, atrial fibrillation, and hypertension, the obstructive sleep apnea syndrome retained a statistically significant association with stroke or death (hazard ratio, 1.97; 95 percent confidence interval, 1.12 to 3.48; P=0.01). In a trend analysis, increased severity of sleep apnea at baseline was associated with an increased risk of the development of the composite end point (P=0.005). conclusions The obstructive sleep apnea syndrome significantly increases the risk of stroke or death from any cause, and the increase is independent of other risk factors, including hypertension. abstract Downloaded from www.nejm.org on January 23, 2010 . Copyright © 2005 Massachusetts Medical Society. All rights reserved. 第 85 页

OBSTRUCTIVE SLEEP APNEA AS A RISK FACTOR FOR STROKE AND DEATH TROKE IS THE SECOND LEADING CAUSE for reasons other than the evaluation of suspected of death worldwide and the leading cause of sleep-disordered breathing(e. g, narcolepsy or long-term disability. 1, 2 Strategies for stroke movement disorder); if they had a history of stroke prevention, including the control of hypertension, myocardial infarction, or tracheostomy; or if the treatment of atrial fibrillation, and smoking cessa- entire polysomnographic study was performed with tion, have reduced the disease burden, but stroke airway pressurization(e. g, continuous positive air- still remains an important public health challenge. way pressure for therapeutic purposes) A better understanding of the risk factors for stroke Participants or their family members gave either is needed to develop additional preventive strategies. written or oral informed consent at the time of The obstructive sleep apnea syndrome is a treat- follow-up ascertainment. The study was approved able form of disordered breathing in which the up- by the Human Investigation Committee at Yale Uni- per airway closes repeatedly during sleep The syn- versity School of Medicine. drome is associated with vascular risk factors and with substantial cardiovascular morbidity and mor- BASELINE ASSESSMENT tality. 3 Several studies have shown a prevalence of Data on demographic characteristics, sleep and the syndrome among patients with stroke that ex- medical history, medication use, and habits were ceeds 60 percent, 4-7 as compared with 4 percent in obtained with the use of a standardized question the middle-aged adult population. 8 naire administered by a trained technologist before Whether the relation between the syndrome and the initiation of overnight polysomnography; the stroke is independent of confounding risk fac- questionnaires were reviewed by a physician. Each tors, such as hypertension, hyperlipidemia, diabetes patient's height and weight were recorded at the mellitus, and smoking, is not clear. Several cross- time of polysomnography and used to calculate the sectional analyses6, 9-14 have shown an increase in body-mass index the risk of stroke with sleep-disordered breathing Sleep-history data included a validated measure that is similar in magnitude to the effect of other of daytime sleepiness (the Epworth Sleepiness cardiovascular risk factors. A study of patients with Scale)5 and self-reported habitual snoring, which acute stroke?demonstrated that obstructive apnea was defined as loud snoring occurring"frequently persisted despite neurologic recovery, suggesting or"constantly "Data regarding medications includ that the obstructive sleep apnea syndrome may have ed the daily use of beta-blockers, angiotensin-con predated the development of stroke. We therefore verting-enzyme inhibitors, other antihypertensive hypothesized that patients with the syndrome have medications, antiplatelet therapy, anticoagulants, an increased risk of stroke or death from any cause oral medications for the treatment of diabetes, in- that is independent of other cerebrovascular risk sulin, and lipid-lowering medications. Risk-factor factors data included a history of hypertension, atrial fibril lation, diabetes mellitus, or hyperlipidemia, either METHODS reported by the patient on the baseline medical questionnaire or noted by the referring physician. STUDY POPULATION In addition, atrial fibrillation on electrocardiogram We conducted an observational cohort study. The phy during polysomnography was considered suf- cohort consisted of patients who were referred to ficient evidence to establish that diagnosis the Yale Center for Sleep Medicine specifically for Patients were classified according to whether the evaluation of sleep-disordered breathing, under- they were current or former smokers or had never went at least two hours of attended sleep monitor- smoked; data were elicited, if applicable, on the ing, completed a 10-page questionnaire on their number of pack-years of smoking. The history of sleep and medical history, and were 50 or more alcohol consumption was based on the average years old. The exposure group was defined a priori number of drinks per day and the number of years as having an apnea-hypopnea index of 5 or higher of drinking ( five with an apnea-hypopnea index of less than 5 con- POLYSOMNOGRAPHY stituted the comparison group Participants underwent attended overnight poly- Patients were excluded if they had been referred sonography with the use of Grass data-acquisition N EnglJ Med 353:9 Www.nejm.org NovembEr 10, 2005 第86页 Downloadedfromwww.nejm.orgonJanuary23,2010.Copyright@2005MassachusettsMedicalSocietyAllrightsreserved

n engl j med 353;19 www.nejm.org november 10, 2005 obstructive sleep apnea as a risk factor for stroke and death 2035 troke is the second leading cause of death worldwide and the leading cause of long-term disability.1,2 Strategies for stroke prevention, including the control of hypertension, treatment of atrial fibrillation, and smoking cessation, have reduced the disease burden, but stroke still remains an important public health challenge. A better understanding of the risk factors for stroke is needed to develop additional preventive strategies. The obstructive sleep apnea syndrome is a treatable form of disordered breathing in which the upper airway closes repeatedly during sleep. The syndrome is associated with vascular risk factors and with substantial cardiovascular morbidity and mortality.3 Several studies have shown a prevalence of the syndrome among patients with stroke that exceeds 60 percent,4-7 as compared with 4 percent in the middle-aged adult population.8 Whether the relation between the syndrome and stroke is independent of confounding risk factors, such as hypertension, hyperlipidemia, diabetes mellitus, and smoking, is not clear. Several crosssectional analyses6,9-14 have shown an increase in the risk of stroke with sleep-disordered breathing that is similar in magnitude to the effect of other cardiovascular risk factors. A study of patients with acute stroke7 demonstrated that obstructive apnea persisted despite neurologic recovery, suggesting that the obstructive sleep apnea syndrome may have predated the development of stroke. We therefore hypothesized that patients with the syndrome have an increased risk of stroke or death from any cause that is independent of other cerebrovascular risk factors. study population We conducted an observational cohort study. The cohort consisted of patients who were referred to the Yale Center for Sleep Medicine specifically for the evaluation of sleep-disordered breathing, underwent at least two hours of attended sleep monitoring, completed a 10-page questionnaire on their sleep and medical history, and were 50 or more years old. The exposure group was defined a priori as having an apnea–hypopnea index of 5 or higher (five or more events per hour of sleep); patients with an apnea–hypopnea index of less than 5 constituted the comparison group. Patients were excluded if they had been referred for reasons other than the evaluation of suspected sleep-disordered breathing (e.g., narcolepsy or movement disorder); if they had a history of stroke, myocardial infarction, or tracheostomy; or if the entire polysomnographic study was performed with airway pressurization (e.g., continuous positive airway pressure for therapeutic purposes). Participants or their family members gave either written or oral informed consent at the time of follow-up ascertainment. The study was approved by the Human Investigation Committee at Yale University School of Medicine. baseline assessment Data on demographic characteristics, sleep and medical history, medication use, and habits were obtained with the use of a standardized questionnaire administered by a trained technologist before the initiation of overnight polysomnography; the questionnaires were reviewed by a physician. Each patient’s height and weight were recorded at the time of polysomnography and used to calculate the body-mass index. Sleep-history data included a validated measure of daytime sleepiness (the Epworth Sleepiness Scale)15 and self-reported habitual snoring, which was defined as loud snoring occurring “frequently” or “constantly.” Data regarding medications included the daily use of beta-blockers, angiotensin-converting–enzyme inhibitors, other antihypertensive medications, antiplatelet therapy, anticoagulants, oral medications for the treatment of diabetes, insulin, and lipid-lowering medications. Risk-factor data included a history of hypertension, atrial fibrillation, diabetes mellitus, or hyperlipidemia, either reported by the patient on the baseline medical questionnaire or noted by the referring physician. In addition, atrial fibrillation on electrocardiography during polysomnography was considered sufficient evidence to establish that diagnosis. Patients were classified according to whether they were current or former smokers or had never smoked; data were elicited, if applicable, on the number of pack-years of smoking. The history of alcohol consumption was based on the average number of drinks per day and the number of years of drinking. polysomnography Participants underwent attended overnight polysomnography with the use of Grass data-acquisition s methods Downloaded from www.nejm.org on January 23, 2010 . Copyright © 2005 Massachusetts Medical Society. All rights reserved. 第 86 页

The NEW ENGLAND JOURNAL Of MEDICINE systems(Astro-Med)on the basis of a protocol de- for sleep apnea in our cohort and an incidence of scribed previously. #A single, attended polysomno- stroke per year of 1.5 percent23 during a four-year graphic study that was conducted during an entire follow-up period, a sample of 840 patients was re- night was used to establish the presence of sleep quired in order to achieve 80 percent power to de- apnea. 16 Sleep stages were scored in 30-second tecta relative risk of 2.0 at the 5 percent(two-tailed) epochs according to standard criteria. 17 Total ces- significance level. sation of airflow at the nose and mouth for at least A series of prespecified time-to-event analyses 10 seconds was classified as apnea(as obstructive were performed to examine the effect of the ob- apnea if respiratory efforts were present and as structive sleep apnea syndrome on the outcome. central apnea if respiratory efforts were absent). The time until the composite end point was taken Partial airway closure, resulting in a diminution of to be the time until stroke, if the patient was con- airflow by more than 30 percent for at least 10 sec- tacted and found to have had a stroke, and the time onds and associated with oxygen desaturation of until death, if the patient was not reached but was 4 percent or more, was termed hypopnea. 18 Cal- found to have died. Data were censored at the time culated polysomnographic variables included the of the contact if the patient was reached and was apnea-hypopnea index and the arousal index(the found not to have had a stroke and were censored number of arousals per hour of sleep at day 1 if the patient was not reached (or unwilling to be contacted and was not found to have died OUTCOMES The Kaplan-Meier method and the log-rank test Each patient was sent a follow-up questionnaire, were used to compare event-free survival among pa- which included questions regarding current state tients with and those without the obstructive sleep of health, occurrence of stroke, hospitalizations, apnea syndrome. With the use of proportional- and treatment of sleep disorders since the baseline hazards analysis, hazard ratios and 95 percent con assessment Strokes and transient ischemic attacks fidence intervals were generated for the unadjusted (TIA)were ascertained with the use of a validated association between sleep-apnea status or other questionnaire designed to bea practical and reliable baseline characteristics and the end point of stroke means of ascertaining stroke status. 19 Attempts or death from any cause Hazard ratios were then were made to telephone patients who had not re- adjusted for the confounding effects of other base- sponded to the initial mailed questionnaire. Family line characteristics, including age, sex, race, smok- members were asked to provide information about ing status, alcohol-consumption status, body-mass patients who were not able to participate owing to index, and the presenceorabsence of diabetes mel death. illness, or dementia. litus, hyperlipidemia, atrial fibrillation, and hyper- A physician investigator who was unaware of the tension. Because of the possibility that controlling patient's status with regard to the obstructive sleep for hypertension could constitute "overadjustment" apnea syndromevalidated reported strokes and TIAs (i.e, accounting for a variable on the causal path by reviewing medical records. Diagnoses were de- way), 24 models were created both with and without termined according to criteria of the National Insti- theinclusion ofhypertension. Finally, a trend analy tute of Neurological Disorders and Stroke 20 for the sis, with the use of the chi-square test for linear classification of cerebrovascular events. The exact trend, was performed to analyze whether an in date of the stroke or TIA was recorded. Vital records creased severity of the obstructive sleep apnea from the Connecticut Department of Public Health syndrome(on the basis of quartiles of the apnea- (regarding in-state deaths)and the Social Security hypopnea index) was associated with an increased Administration Death Master File 21, 22(regarding risk of stroke or death from any cause. out-of-state deaths )were used to determine or con- Students t-test was used an val firm death. The exact date of death was recorded. ues at baseline among patients who had the obstruc tive sleep apnea syndrome with those in the compar STATISTICAL ANALYSIS on group. Categorical data were compared with The primary outcome was the composite end point the use of the chi-square test. All statistical tests of incident stroke (including TIA, which hereafter were performed with the use of SAS software(SAS will be reported as stroke)or death from any cause. Institute). All reported P values are two-sided, and With the assumption of a prevalence of 60 percent no interim analyses were conducted. S-plus soft NenGljMed353:9www.nejm.orgNovemBer1o,2005 第87页 Downloadedfromwww.nejm.orgonJanuary23,2010.Copyright@2005MassachusettsMedicalSocietyAllrightsreserved

n engl j med 353;19 www.nejm.org november 10, 2005 The new england journal of medicine 2036 systems (Astro-Med) on the basis of a protocol described previously.4 A single, attended polysomnographic study that was conducted during an entire night was used to establish the presence of sleep apnea.16 Sleep stages were scored in 30-second epochs according to standard criteria.17 Total cessation of airflow at the nose and mouth for at least 10 seconds was classified as apnea (as obstructive apnea if respiratory efforts were present and as central apnea if respiratory efforts were absent). Partial airway closure, resulting in a diminution of airflow by more than 30 percent for at least 10 seconds and associated with oxygen desaturation of 4 percent or more, was termed hypopnea.18 Calculated polysomnographic variables included the apnea–hypopnea index and the arousal index (the number of arousals per hour of sleep). outcomes Each patient was sent a follow-up questionnaire, which included questions regarding current state of health, occurrence of stroke, hospitalizations, and treatment of sleep disorders since the baseline assessment. Strokes and transient ischemic attacks (TIA) were ascertained with the use of a validated questionnaire designed to be a practical and reliable means of ascertaining stroke status.19 Attempts were made to telephone patients who had not responded to the initial mailed questionnaire. Family members were asked to provide information about patients who were not able to participate owing to death, illness, or dementia. A physician investigator who was unaware of the patient’s status with regard to the obstructive sleep apnea syndrome validated reported strokes and TIAs by reviewing medical records. Diagnoses were determined according to criteria of the National Institute of Neurological Disorders and Stroke20 for the classification of cerebrovascular events. The exact date of the stroke or TIA was recorded. Vital records from the Connecticut Department of Public Health (regarding in-state deaths) and the Social Security Administration Death Master File21,22 (regarding out-of-state deaths) were used to determine or confirm death. The exact date of death was recorded. statistical analysis The primary outcome was the composite end point of incident stroke (including TIA, which hereafter will be reported as stroke) or death from any cause. With the assumption of a prevalence of 60 percent for sleep apnea in our cohort and an incidence of stroke per year of 1.5 percent23 during a four-year follow-up period, a sample of 840 patients was required in order to achieve 80 percent power to detect a relative risk of 2.0 at the 5 percent (two-tailed) significance level. A series of prespecified time-to-event analyses were performed to examine the effect of the obstructive sleep apnea syndrome on the outcome. The time until the composite end point was taken to be the time until stroke, if the patient was contacted and found to have had a stroke, and the time until death, if the patient was not reached but was found to have died. Data were censored at the time of the contact if the patient was reached and was found not to have had a stroke and were censored at day 1 if the patient was not reached (or unwilling to be contacted) and was not found to have died. The Kaplan–Meier method and the log-rank test were used to compare event-free survival among patients with and those without the obstructive sleep apnea syndrome. With the use of proportionalhazards analysis, hazard ratios and 95 percent confidence intervals were generated for the unadjusted association between sleep-apnea status or other baseline characteristics and the end point of stroke or death from any cause. Hazard ratios were then adjusted for the confounding effects of other baseline characteristics, including age, sex, race, smoking status, alcohol-consumption status, body-mass index, and the presence or absence of diabetes mellitus, hyperlipidemia, atrial fibrillation, and hypertension. Because of the possibility that controlling for hypertension could constitute “overadjustment” (i.e., accounting for a variable on the causal pathway),24 models were created both with and without the inclusion of hypertension. Finally, a trend analysis, with the use of the chi-square test for linear trend, was performed to analyze whether an increased severity of the obstructive sleep apnea syndrome (on the basis of quartiles of the apnea– hypopnea index) was associated with an increased risk of stroke or death from any cause. Student’s t-test was used to compare mean values at baseline among patients who had the obstructive sleep apnea syndrome with those in the comparison group. Categorical data were compared with the use of the chi-square test. All statistical tests were performed with the use of SAS software (SAS Institute). All reported P values are two-sided, and no interim analyses were conducted. S-Plus softDownloaded from www.nejm.org on January 23, 2010 . Copyright © 2005 Massachusetts Medical Society. All rights reserved. 第 87 页