- Email: [email protected]
Research committee
t The Research Committee of APK wishes to foster quality and innovative research among its membership. Innumerable questions regarding even the most fundamental aspects of infection control and epidemiology remain to be answered. The process of designing a research study is critical to the utility of its results but may be intimidating to practitioners who have had little formal training or experience in fashioning a properly crafted research project. In this section we provide a brief overview of the initial steps involved in planning a research project. A few of the more common types of study design and their characteristics are set down in outline form to familiarize the reader with the choices available. Each type of design is suited to different tasks and has its own strengths and weaknesses. A bibliography is presented, and its use is essential for learning how to pick and choose among the study types. The initial step for the researcher is, of course, to choose a problem to study and then outline specific questions to be answered. The list of suggested infection control research topics provided by the Research Committee would be a good starting point (AMERICAN .T~URNAL OF INFECTION CONTROL, Qctober l985;13[5]:31.&3A, and October 1986;14[5]:26A-SOA). Details about the project and specific definitions regarding end points should be meticulously drafted in written form at this stage. A literature review is required at this juncture as well. All previous studies that have examined the problem should be surveyed to prevent a duplication of effort. An additional benefit of a thorough literature review includes tips on solving problems that previous investigators in the field have noted. The investigator is then faced with a choice of which study design is best suited to answer Certhe particular needs of his or her project. tain common types of study design are outlined below. Proper study of the bibliography accompanying this outline should give the novice re-
searcher a thorough groun method.s and study design.
ing
:i research
I. ~~~dQ~~~~d clinical trials by chance me&aAIlocation of t~eatrn~~t nism to control ias. Adv~~~~~~~ c&t owe&d designs t eliminate bias su as selection bias, time (period) bias, observation bias A. Parallel designs of ~~~~e~~~ are 1 e Two or more groups treated separately but co~~~r~~~Iy as part of the same stu 2. Factors to consider: a. Entry ~-~quireme~ts. sampling from well-de&n. ulation. selection criteria well described. b. ~~~~~O~i~~~ ~O~~~i~~~ im” Absence of statisticai d of covariates groups is not
bet\nreen t~~~~rne~~ necessarily evidence that the groups are similar.
(1) Randomization (2) Stratification with Mocking (3) Adjustment: use of ~~ig~t~daverage of results in v~~i~~s strata I Crossover designs Each patient receives two or more :.reatments in sequence, and ~~~~o~les in the same patient are contrasted. 1. Advantage: smailer sample can psovide statistical accuracy af a large parallel study eisefd in short 2. Ideal and especially treatments with few extr.eme side fects and in ~cornrn~~ diseases
3. Factors to consider: a, Carryover and
period
ef-
effects.
Volume June
15 Number
3
Committee reports 31 A
1987
Therapeutic or deleterious effect of first therapy may persist in second phase. Watch for “regression toward mean,” b. Crossover rules and timing should be predetermined and time dependent rather than disease-state dependent. When possible, crossover should be blinded to patient and observer. c. Dropouts. Effects are more pronounced in crossover than in parallel studies. d. Treatment sequence should be randomized. e. Statistical analysis may require multivariate methods. Advantages of parallel design over crossover designs: I. Allows more straightforward analysis 2. Dependent on assumptions about disease process 3. Usually lower dropout rate 4. Greater precision Exceptions to randomized controlled
trials: 1. Epidemiologic research of toxins or infectious agents as cause of disease 2. Rare diseases 3 f Self-controlled sludies II. Epidemiologic research Epidemiologic studies of causes of disease depend on observational procedures because noxious or toxic agents cannot be tested experimentally in people. Types of observational studies: A. Cohort (prospective or forward directed) Exposed population received the noxious or toxic agent through mechanisms other than random assignment (selfimposed, medication, environmental pollutant, or toxin). Both exposed and unexposed (control) populations are followed and compared and the following calculated: I. Date of occurrence 2. Attributable risk or risk ratio or relative
risk Problems: costly in time and money, requires a large population to study
B. Case-control (retrospective) Studies performed in people who already have the disease and compared with a control group from people in whom the disease has not been noted. The two groups are followed backward with the aim of determining each member’s antecedent exposure to the suspected etiologic agent. Advantage: does not require large sample size or prolonged follow-up periods Problems with epidemiologic research: 1. Quality control a. Verification of basic groups b. Confirmation of subsequent data Cross-checking Repeated interviews or questionnaires Checking subject’s memory of distant events 2. Unfair comparisons (covariate imbalance) a. Susceptibility bias (“protopathic bias”) b. Detection bias c. Exclusion or migration bias 3. Validity or cogency of methods a. Methodologic vigilance Berkson’s bias b. Retroactive alteration of research hypothesis (1) Changes in suspected agents (2) Analysis of multiple subgroups (3) Changes in control groups What to do: 1. To reduce migration and detection
bias * Use the same eligibility criteria and subsequent surveillance procedures that would have been used in randomized studies ~ 2. To reduce susceptibility bias l Aside from demographic factors, obtain all necessary or pertinent clinical information, for example: a. Psychological factors b. Parental longevity c. Health status d. Reasons for exposure
Committee reports 3.
improve collection and confwmation of data e Careful selection and verification of diagnosis to avoid both false negatives and false positives 4. To improve cogency of methods @ If alteration of agents of control is done, results should be interpreted cautiously-not conclusions, but tentative hypothesis requiring further study. * atch for Be&son’s bias and avoid. Adapted Research
Ts
from: Teaching notes. Laboratory. Eufronio
Hartford Maderazo,
Hospital M.D.,
Medical director.
eferen 1. Spodick DH. The randomized controlied &nicai sriai. Am J Med 1982;-?3:420-5. 2. Louis TA, Lavori PW, Bailar JC 3d, Pcdlar,sky M. Crossover and self-controlled designs in clinical resua;rch. N Engl J Med 1984;310:24-31. 3. Riegelman RK. Studying a study and testirig a test. Boston: Little, Brown & Co., 1981. 4. Zelen M. A new design for randomized’ciiaicaf: trials. N Engl J Med 1979;300:1242-5. 5. Lavori PW, Louis TA, Bailar JC 3d, Poiansky M. Des~igns for experiments-.paraIlel comparisons of tYeatment.
N Engl J Med 1983;309%!91-9. 5. Feinstein AR, methods, and 1982;307:1611-7.
Horwitz RL epidemiologic
Double standards, scientific research. N Engi J Med
1987 Research Committee rian Cooper, M.D. FAAN Deborah Coleman,
Answer
R.N., MS., chairman
sheet in this issue for April 1987 Continuing
The answer
Educatio
sheet for the Continuing Education examination OII the article “5~$rylococcu.s epidemidis: A Significant Nosocomial Pathogen” was inadvertently omitted from the April issue. An additional answer sheet has been provided in this issue for that examination, and the deadline has been extended by 2 months. Answer forms must now be postmarked by Dec. 25, 1987. Please note that this issue contains two answer sheets, one on p. 38W for the examination in this issue on “‘Cytomegalovirus: Epidemiology and h&ction Control” and one on p. 39A for the April test. Please make certain that you use the correct form for the examination that you are taking.
I
[ ; 1 g ! 3
1
searcher a thorough groun method.s and study design.
ing
:i research
I. ~~~dQ~~~~d clinical trials by chance me&aAIlocation of t~eatrn~~t nism to control ias. Adv~~~~~~~ c&t owe&d designs t eliminate bias su as selection bias, time (period) bias, observation bias A. Parallel designs of ~~~~e~~~ are 1 e Two or more groups treated separately but co~~~r~~~Iy as part of the same stu 2. Factors to consider: a. Entry ~-~quireme~ts. sampling from well-de&n. ulation. selection criteria well described. b. ~~~~~O~i~~~ ~O~~~i~~~ im” Absence of statisticai d of covariates groups is not
bet\nreen t~~~~rne~~ necessarily evidence that the groups are similar.
(1) Randomization (2) Stratification with Mocking (3) Adjustment: use of ~~ig~t~daverage of results in v~~i~~s strata I Crossover designs Each patient receives two or more :.reatments in sequence, and ~~~~o~les in the same patient are contrasted. 1. Advantage: smailer sample can psovide statistical accuracy af a large parallel study eisefd in short 2. Ideal and especially treatments with few extr.eme side fects and in ~cornrn~~ diseases
3. Factors to consider: a, Carryover and
period
ef-
effects.
Volume June
15 Number
3
Committee reports 31 A
1987
Therapeutic or deleterious effect of first therapy may persist in second phase. Watch for “regression toward mean,” b. Crossover rules and timing should be predetermined and time dependent rather than disease-state dependent. When possible, crossover should be blinded to patient and observer. c. Dropouts. Effects are more pronounced in crossover than in parallel studies. d. Treatment sequence should be randomized. e. Statistical analysis may require multivariate methods. Advantages of parallel design over crossover designs: I. Allows more straightforward analysis 2. Dependent on assumptions about disease process 3. Usually lower dropout rate 4. Greater precision Exceptions to randomized controlled
trials: 1. Epidemiologic research of toxins or infectious agents as cause of disease 2. Rare diseases 3 f Self-controlled sludies II. Epidemiologic research Epidemiologic studies of causes of disease depend on observational procedures because noxious or toxic agents cannot be tested experimentally in people. Types of observational studies: A. Cohort (prospective or forward directed) Exposed population received the noxious or toxic agent through mechanisms other than random assignment (selfimposed, medication, environmental pollutant, or toxin). Both exposed and unexposed (control) populations are followed and compared and the following calculated: I. Date of occurrence 2. Attributable risk or risk ratio or relative
risk Problems: costly in time and money, requires a large population to study
B. Case-control (retrospective) Studies performed in people who already have the disease and compared with a control group from people in whom the disease has not been noted. The two groups are followed backward with the aim of determining each member’s antecedent exposure to the suspected etiologic agent. Advantage: does not require large sample size or prolonged follow-up periods Problems with epidemiologic research: 1. Quality control a. Verification of basic groups b. Confirmation of subsequent data Cross-checking Repeated interviews or questionnaires Checking subject’s memory of distant events 2. Unfair comparisons (covariate imbalance) a. Susceptibility bias (“protopathic bias”) b. Detection bias c. Exclusion or migration bias 3. Validity or cogency of methods a. Methodologic vigilance Berkson’s bias b. Retroactive alteration of research hypothesis (1) Changes in suspected agents (2) Analysis of multiple subgroups (3) Changes in control groups What to do: 1. To reduce migration and detection
bias * Use the same eligibility criteria and subsequent surveillance procedures that would have been used in randomized studies ~ 2. To reduce susceptibility bias l Aside from demographic factors, obtain all necessary or pertinent clinical information, for example: a. Psychological factors b. Parental longevity c. Health status d. Reasons for exposure
Committee reports 3.
improve collection and confwmation of data e Careful selection and verification of diagnosis to avoid both false negatives and false positives 4. To improve cogency of methods @ If alteration of agents of control is done, results should be interpreted cautiously-not conclusions, but tentative hypothesis requiring further study. * atch for Be&son’s bias and avoid. Adapted Research
Ts
from: Teaching notes. Laboratory. Eufronio
Hartford Maderazo,
Hospital M.D.,
Medical director.
eferen 1. Spodick DH. The randomized controlied &nicai sriai. Am J Med 1982;-?3:420-5. 2. Louis TA, Lavori PW, Bailar JC 3d, Pcdlar,sky M. Crossover and self-controlled designs in clinical resua;rch. N Engl J Med 1984;310:24-31. 3. Riegelman RK. Studying a study and testirig a test. Boston: Little, Brown & Co., 1981. 4. Zelen M. A new design for randomized’ciiaicaf: trials. N Engl J Med 1979;300:1242-5. 5. Lavori PW, Louis TA, Bailar JC 3d, Poiansky M. Des~igns for experiments-.paraIlel comparisons of tYeatment.
N Engl J Med 1983;309%!91-9. 5. Feinstein AR, methods, and 1982;307:1611-7.
Horwitz RL epidemiologic
Double standards, scientific research. N Engi J Med
1987 Research Committee rian Cooper, M.D. FAAN Deborah Coleman,
Answer
R.N., MS., chairman
sheet in this issue for April 1987 Continuing
The answer
Educatio
sheet for the Continuing Education examination OII the article “5~$rylococcu.s epidemidis: A Significant Nosocomial Pathogen” was inadvertently omitted from the April issue. An additional answer sheet has been provided in this issue for that examination, and the deadline has been extended by 2 months. Answer forms must now be postmarked by Dec. 25, 1987. Please note that this issue contains two answer sheets, one on p. 38W for the examination in this issue on “‘Cytomegalovirus: Epidemiology and h&ction Control” and one on p. 39A for the April test. Please make certain that you use the correct form for the examination that you are taking.
I
[ ; 1 g ! 3
1