DO-oog- - UPV/EHU · DO-oog-Xiaohan Hu, PhD, Jarnes G. Wright, MD, MPh,a Robin S. McLeod, MD, Al...

4
DO-oog- Xiaohan Hu, PhD, Jarnes G. Wright, MD, MPh,a Robin S. McLeod, MD, Al Lossing, MD, and Beverly C. Walters, MD, Taranta, Ontaria, Canada From the Clinical EPidemiologyGroup, Depaltment 01 Sulgel)', Univl'1"Sity 01 Toronto, Ontario, Ganada CLI~I(:AI, EPIDEMIOLOGY ASA DISCIPLINE hasevolved rap- idly during the last decade, However, application of ep- idemiologic principIes in surgical clinical research has, been limited. In the March 1996 issue of SURGERY we discussed the Arinciples, application, andlimitations of randomized cÍin/cal trials (RCTs) in surgery. RCTs are often \1ewed a$ the gold standard for evaluation of therapeuti¿ efficacy because they are, the closest that clinical medicine can come to the conu'olled en\1rOn- ment oflaboratol"y research, Ho\vever, clinical decisions often have to be made in the absence of data obtained this \vay, In this issue we ,viII discuss some alternatives that can be used in surgical clinical research. Alternati,es to RCTs fall into the pur\1ewof observa- tional epidemiology, which can be grouped into five majar categories: case reports, case series, cross-sec- tional studies, case-control studies. and cohor.t studies. The former three are often referred to as descriptive studies, and the remaining t\vo are labeled as ana~ytic or comparative studies, DES CRIPTIVE STUDlES Descripti,estudies differ fromanalytic studies in that they contain no control or comparison groups and are usually lmdertaken ,vhen little is known about the epi- demiolog)' of the disease in qllestion. The anal}'Sisoften focllses on a descliption of the disease in the Stlldy pop- 1.uation according to patient charatteristics such as age, gender, ethnic origin, socioeconomic class, occupation, geographic afea, time of occurrence, and clinical course. The results of descriptive studies are often used for hy- pothesis generation, baseline data, rationale for sample size estimation, and, in the absence oftreatn1ent, a de- scription ofthe natural history of the disease. Descrip- tive studies are also used in fue early.stagesof the e\'al- uation of a new treatment. They are generally consid- ered as the first, but sometimes necessaf}',step to\\'ard subsequent analytic studies or RCTs. Case reporto A case report is the description of clin- ical events of one or severalpatients in a narrative folm. Its place in !;1inical research lies in reporting rafe and significantcomplications of disease, or describing the u'eatment of an un usual disease.The purpose of a case report is to prompt colleagues to lookfor similar effects in theirpatients, so tbat initial isolated obser\'ations can be either verified or nullified. CaSe series. Case series describe the spectrum of clinical features of agroup of patients who are moni- tored from the sanie inception point of the clinical course of a disease. Routinely collected data, such as medical records, are commonly used. \\7l1ena case se- lies is used to describe the response to treatment, patients' clinical courses are usually compared \vith his- torical controls because the study design lacks a com- parative (control) group. This form ofresearch isprone to bias beca use the characteristicsof the patients or the methods of their assessmentmar account for differ- ences in outcome rather than the treatment. As dis- cussed in the pre\ious article inthe March 1996 issueof SVRGERY, caseseries seldom provide proof of therapeu- tic efficacy. Cross-sectional studies. In cross-sectionalstudies in- di\iduals are charactelized by h}pothesized risk factors and a diseaseoC interest at one specific point in time. Acceptcd for publicaüon April 18, 1995. Reprint reques[S:James G. Wright, MD, MPh, S107, The Hospital for Sick Children, 555 Uni\ersity A",., Toranto, Ontarlo ~15G lX8. Canada. aRecipient of a Medical Research Council Scholarship. St.RGERY 1996;1 19:47g.5. Cop)Tight ~ 1996 by ~losb}.-Y"ar Bóok, lnc, 0039-6060/96/55.00 + O 11/55/70912 SURGERY 473

Transcript of DO-oog- - UPV/EHU · DO-oog-Xiaohan Hu, PhD, Jarnes G. Wright, MD, MPh,a Robin S. McLeod, MD, Al...

DO-oog-

Xiaohan Hu, PhD, Jarnes G. Wright, MD, MPh,a Robin S. McLeod, MD, Al Lossing, MD,and Beverly C. Walters, MD, Taranta, Ontaria, Canada

From the Clinical EPidemiology Group, Depaltment 01 Sulgel)', Univl'1"Sity 01 Toronto, Ontario,Ganada

CLI~I(:AI, EPIDEMIOLOGY AS A DISCIPLINE hasevolved rap-

idly during the last decade, However, application of ep-

idemiologic principIes in surgical clinical research has,

been limited. In the March 1996 issue of SURGERY we

discussed the Arinciples, application, andlimitations ofrandomized cÍin/cal trials (RCTs) in surgery. RCTs are

often \1ewed a$ the gold standard for evaluation of

therapeuti¿ efficacy because they are, the closest that

clinical medicine can come to the conu'olled en\1rOn-

ment oflaboratol"y research, Ho\vever, clinical decisions

often have to be made in the absence of data obtained

this \vay, In this issue we ,viII discuss some alternatives

that can be used in surgical clinical research.

Alternati,es to RCTs fall into the pur\1ewof observa-

tional epidemiology, which can be grouped into five

majar categories: case reports, case series, cross-sec-

tional studies, case-control studies. and cohor.t studies.

The former three are often referred to as descriptivestudies, and the remaining t\vo are labeled as ana~ytic or

comparative studies,

DES CRIPTIVE STUDlES

Descripti,estudies differ fromanalytic studies in that

they contain no control or comparison groups and are

usually lmdertaken ,vhen little is known about the epi-

demiolog)' of the disease in qllestion. The anal}'Sisoften

focllses on a descliption of the disease in the Stlldy pop-

1.uation according to patient charatteristics such as age,

gender, ethnic origin, socioeconomic class, occupation,geographic afea, time of occurrence, and clinical course.The results of descriptive studies are often used for hy-pothesis generation, baseline data, rationale for samplesize estimation, and, in the absence oftreatn1ent, a de-scription ofthe natural history of the disease. Descrip-tive studies are also used in fue early.stages of the e\'al-uation of a new treatment. They are generally consid-ered as the first, but sometimes necessaf}', step to\\'ardsubsequent analytic studies or RCTs.

Case reporto A case report is the description of clin-ical events of one or several patients in a narrative folm.Its place in !;1inical research lies in reporting rafe andsignificantcomplications of disease, or describing theu'eatment of an un usual disease. The purpose of a casereport is to prompt colleagues to lookfor similar effectsin theirpatients, so tbat initial isolated obser\'ations canbe either verified or nullified.

CaSe series. Case series describe the spectrum ofclinical features of agroup of patients who are moni-tored from the sanie inception point of the clinicalcourse of a disease. Routinely collected data, such asmedical records, are commonly used. \\7l1en a case se-lies is used to describe the response to treatment,patients' clinical courses are usually compared \vith his-torical controls because the study design lacks a com-parative (control) group. This form ofresearch isproneto bias beca use the characteristicsof the patients or themethods of their assessment mar account for differ-ences in outcome rather than the treatment. As dis-cussed in the pre\ious article inthe March 1996 issue ofSVRGERY, case series seldom provide proof of therapeu-tic efficacy.

Cross-sectional studies. In cross-sectional studies in-di\iduals are charactelized by h}pothesized risk factorsand a diseaseoC interest at one specific point in time.

Acceptcd for publicaüon April 18, 1995.

Reprint reques[S:James G. Wright, MD, MPh, S107, The Hospital forSick Children, 555 Uni\ersity A",., Toranto, Ontarlo ~15G lX8. Canada.

aRecipient of a Medical Research Council Scholarship.

St.RGERY 1996;1 19:47g.5.

Cop)Tight ~ 1996 by ~losb}.-Y"ar Bóok, lnc,

0039-6060/96/55.00 + O 11/55/70912

SURGERY 473

SurgeryApril1996

474 Hu el al.

tion of appropriate controls is probably the most diffi-cult and c~ntroversial aspect of a case-control study. InprincipIe fue control group is intended to proVide anestimate of fue exposure rate that w.ould be expected tooccur in the cases ifno association was present benveenfue study disease and exposure.3 Consequently, patien tswith conditions known to predispose to or against theexposure under study should be excluded from fuecontrol group. For exampl~, patients with lung cancershould be excluded as controls in a study of associationbetween smoking and coronary heart disease. Other-wise, the risk factor (smoking) ,viII be underestimatedbecause subjects with a past smoking historyare over-represented in fue controlgroup asa resultofthe causalassociation benveen lung cancer and smoking. Simi-larly, in astudy of the effectiveness of Pap smear tests inreducing mortality from cen;cal cancer,patients "ith ahysterectomy should be exclllded because they,vouldbe less likely to have a Pap smear test, ,vhich could re-duce the apparent effectiveness of the test (you ,voltldexpect a ~uch higherpercentage of,vomen in controlgroup to have Pap smear test if the test is highly effec-tive). In clinical research, cases areusually selected froma medical institute and fue control grOllp is composedofpatients ,vith other diseases-conditions from the sameinstitute.

Information bias occurs as a results of flaws in themethods used to collect the data. For example, patients,vith breast cancer might recall exposures more thor-oughly than their controls. To prevent informationbiasit is imperative that efforts to ascertain exposure areco~paraple in the two groups. Whenever possible, forexample, intelViewers should be 1.ma\vare or blind todisease status of study subjects, and patients or medicalrecord abstractors should not kno,v the main purposeof fue study.

Case-control studies can also be used toevaluate theeffectiveness of treatments but are rarely used in surgi-cal clinical research.4 For example, Nissen and Toupetfundoplication procedur~s are conc1.lrrently practicedin our hospital as altemati:ves for children \\;th gastroe-sophageal reflux (GER) for whom medical manage-ment fails. At present fue choice ben,.een fue nvo pro-cedures is based on surgeon preference. Assumingthatsurgical skills are comparable among the surgeons, acase-control study could be conducted to determine

.-whetherone procedure is superior to theother. Patientsin whom recurrence of reflux has occurred (cases)would be compared ,vith patieuts in whom resolution ofreflux has been observed (controls). The type of fueprocedure received by subjects, along "ith characteris-tics knO\Vfi to influence GER, could be compared. In theabsence of systematic bias the outcome as estimated byodds ratio would reflect the magnitude of the relativetherapeutic effectiveness benv~en the rlvo procedures.

Cross-sectional studies are used to assess the magnitude.of a disease in the population and to determin~ the

prevalence of risk factors for disease. One majar advan-tage of a cros~sectional study is that it can be carried outin a timely fashion. Lack of appropriate controls anduncertainty about whether riskfactors precede fue on-set of disease are the two majar threats to the validity ofresults. For example, to explorerisk factors for multipleorgan failure (MOF) inpatients inintensive care unit,a cros~sectional study could be used. AlI patients withthe signs of MOF could have multiple tests includingblood cultures. If the results showed that 50% ofthe pa-tients have positive blood cultures, investigators mightconclude that infection is the origin of MOF andrecommend prophylactic antibiotics to prevent MOF.Inclusion of a control group without MOF, however,might show that 50% ofpatients without MOF algo havepositive blood cultures. Alternatively, infection mar bean early manifestation rather than a cause ofMOF. Thuscross-sectional studies can suggest an association but donot provide proof of causation. &

ANALYTIC STUDIES

Case-control studies. Case-control studies have acontrol or comparison group, butpatients are assem-bled according to the presence or absence of the out-come. In earlier literature case-control studies have'sometimes been referred to as case-referent studies,case-comparison studies, or simply retrospective studies,A group of subjects with a known outcome (cases) anda group o(subjects without the outcome (controls) arecompared for the proportion ofrisk factors.For exam-pIe, in a study of the originofbreastcancer agroup ofwomen who are recently diagnosed with bréast cancer(cases) and a group of women who are free of breastcancer (controls) are recruited. The two groups areco~pared ,vith respect to previous olal contraceptiveuse, estrogen-replacement therapy use, age atfirst birth,menarche, and menopause, history of benign breastdisease or breast feeding, breast cancer in sister ormother, etc. A difference in the frequency ofthese fac-tors in the t\vo groups shows an associátion, which maror mar not be etiologic.

Case-control studies are particularly valuable in etio-logic studies because such design allows comparison ofmultiple factors as illustratedin the above example. Inaddition, they generally can be conducted in a short pe-riod oftime, require small sample sizes, and are the bestdesign for rafe events or late outcomes. Case-controlstudies are superior to descriptive studies because theyhave a control group but are susceptible to other formsof bias.l. 2

The two most prominent types of bias are selectionand information bias. Selection bias results from inap-propriate selection of the control subjects. Identifica-

Hu el al. 475Surge1)Volume 119, Number 4

Cohort study. Cohort studies start by ass~mbling nvogrQups of subjects. One group of patients receives atreatment (or has a risk factor), and the other groupdoes not have the treatment (or the risk factor). Bothgroups are followed up over a period oftime, and thenfue incidence of events in each group is observed andcompared. For example, a group of patients \vith highcholesterol level and a group of control subjects \\1thnormal cholesterollevel are followed up ayer a certainperiodo At the end of study the incidence of coronaryheart disease in each group is tabulated and the rates arecompared. In another example, to compare the effec-tiveness of Nissen and Toupet fundoplication proce-dures for patients \vith intractable GER, patients aregrollped by the procedure they received. Patients areevahlated 2 years after operation, and the olltcomes inthe nvo groups are compared.

Compared \\1th case-conU"ol studies, the majar ad-\'antages of the cohort study are that mllltiple outcomescan be evaluated, rates of olltcomes can be directly cal-culated, and f;xposure al\\'ays precedes the outcome.Moreover, the cohort study is inulitively appcaling be-cause its temporal sequence represcnts the naturalcallsal path\\.ay. Cohort studies are llseftll for evahlationfor risk factors because patients cannot be l-andomly as-signed to risk factor or notto receive the risk factor. Co-hort studies like RCT, however, are not feasible whenthe incidence ofthe event under study is low, such as therisk of fue human immlmodeficiency virus infection afteran infected sharp injury among sllrgeons, or fue time fromexposure to fue presentation of fue event is protracted,SllCh as fue association of smoking \vith emph}'Sema.

One strategy often used to overcome the disadvan-tage of delayed olltcome is to assemble a historical co-hort, in which a group ofindi\1duals is identified on fuebasis of their exposure-intervention status in the pastand then their sllbsequent disease experience is de ter-mined IIp toa point: in the more recentpast, thepresenttime, or even the futllre. This method is termed a his-torical cohort stud}" or retrospective cohort Stlldy. A ret-rospective cohort stlldy shares the advantages of acohort study. In addition, becallse the stlldy can becompleted ina mllcl;¡ shorter time period, it is thereforeconsiderably less expensive. However, such a study isonly slütable when comprehensive information onbaseline StatllS, exposllre, or intel~.ention is available.

The majar source of bias in cohort studies, calledconfounding, derives from imbalance of kno\vo or un-kno\vo prognostic factors between compared groupsoFor example, in a comparison of mortality rates forwomen mth breast cancer \vho are treated \vith eitherlocal exc;:ision or radical mastectomy, tumor size anddifferential levels should be considered as potentialconfounding factors beca use women undergoing radi-cal mastectomy may have a larger size of tumor or thetumor is less differentiated, \vhich could bias results infavor of local excision. To control this confoundingfactor we could (1) confine the stlldy to subjects \vitha narro\v range of tumor size and the same differen-tiallevel, (2) select subjects in such a \,.ay that the dis-triblltion of tumor size and differential levels arecomparable in nvo grollps, or (3) adjllst for tumor sizeand differential levels in the statistical analysiso Thesestrategies, ho\vever, cannot adjust for unkno\vo prog-nostic fa,ctors, \vhich is the rationale for randomizeduoials.

Analytic studies have been a majar thrllst in poplua-tion epidemiology. Ho\,Oe,Oer, their role in clinical re-search has been somewhat dampened probably becallseof the stance taken by some clinkal epidemiologists thatonly RCTs are \'aluable,5 even thOllgh a contrary viewhas been presented by otherso6 The fact that many clin-ical situations can only be evaluated by using observa-tional studies, albeit more susceptible to bias and errors,underlines the r:teed for surgeons to bec~me morefamiliar \vith the alternatives to RCTso UnderstandingprincipIes of observatio~al epidemiology \vill not onlyhelp us to exercise critical appraisal of research reportsblltalso gtlide {lS to design and execute studies that aredestined to be scrutinized by Ollr peers.

REFEREN CF$l. Feinstein AR. Clinical epidemiology. the architeculre of clinical

research.Philadelphia: \\'8 Saunders Company, 1985.2. Sackett DL, Ha}'T1es RB; Tu~"ell P. Clinical epidemiolOg}': a ba-

sic science for clinical medicine. Boston: Little, Bro"'T1, 1985.3. Schlesselmanll. Case-control studies. Oxford, United Kingdom:

Oxford Uni\"ersit}. Press, 1982.4. Solomon MJ. McLeod RS. Clinical studies in surgicaljournals--

ha\"e ,,'e impro\"ed? Dis Colon Rectum 1993;36:43-8.5. Sacks H, Chalmers TC, SmithH. Randomized \"ersus historical

controls for clinical trials: ArnJ Med 1982;72:233-9.6. Van der Linden \\'. Piúalls in randomized surgical triaIs. Surgery

1980;87:258-62.

472 Hall et al. SurgeryApril 1996

34. Lavori PW. LouisTA. Bailar JC. Polansky M. Designs for exper-imenls-parallel comparisons of treaunent. N Engl J Med 1983;

309:1291-9.35. Brennan P, Croft P. Interpreting the resultS of obsen-ational

research: chance Is not such afine thing. BMJ 1994;309:727-30.

36. Chalmers TC. Smith HJr. Blackbum B. et al. A methodfor as-sessing)he quality of a randomized control mal. Control ClinTrials 1981 ;2:31-49.

37. Sackett DL. Cook RJ. Understanding mals. '\'hat measures of ef-ficacy should joumal articles pro\Óde busy clinicians? BMJ 1994;309:755-6.

38. Bradford Hill A The continuing unethical use of placero con-

trols. N EnglJMed 1994;331:394-8.

li errors in the Australian medical literature. Aust NZ J Med

1982;12:7-9.28. Evans M. Presentation of manuscripts for publication in the Brit-

ishJournal of Surgery. Br J Surg 1989;76:1311-5.29. Goodman SN, BerlinjA. The us~ ofpredicted confidente inter-

vals when planning experiments and the misuse of power whenimerpreting results. Ann lmern Med 1994;121:200'5C.

30. Diamond GA, Forrester jS. Trials and statistical verdicts: proba-ble grounds for arrea!. Ann lmern Med 1983;98:385-94.

31. Kelly Mj, Wadswor(hj. What price inconclusive trials? Ann Roy

Coll Surg EngI1993;75:145-6.32. Easterbrook Pj, BerlinjA, Gopalan R, Matthews DR. Publication

bias in clinical research. Lancet 1991;337:867-72.33. Sch,\-artz D, Lellouchj. Explanatory and pragmatic attitudes in

trials.j Chron Dis 1967;210:63748.