Document G5QgYn6E4o6w5baqLZk72oonx

Control Selection at Certain Hospitals Prepared by Otto Wong and Rob Schnatter, 5 May 2004 The selection of controls at each hospital is the responsibility of the Clinical Coordinator at the hospital. The Clinical Coordinators were given a copy of the following instruction: Instruction for selecting controls in the AMLINHL case-control study English Selection of controls (1 case - 2 controls) 1" Same gender 2" Age 5 years 3" No disease of the lymph or blood system 4" Closest admission date 5" Same hospital 6" If the case is an inpatient, controls will be inpatients as well. If the case is an outpatient, controls will be outpatients as well. "("( "("("("("("( (1 "("("( -2 "("("() 1" "("([( 2" "("(5"( 3" "( "( [("( "("( [("( 4" "("([(( "("( 5" "("([( 6" "("( [("( "("( [ ( "("( "("("( "("("(" ?????????. ,'????????. " If a control cannot be found, relax age to 10 years" ???????,? ? ?? 1O? " If only one control can be found, we will keeJ) both the case and the one control. ?.?.?.?. .?.?.?.?.?,?.?.?.?.?.?.?.?.?. .?.?.?.?.?. In February 2004, a review of the questionnaire data of the cases and controls was carried out. Two irregularities were noted: 1" Quite a few controls at the Tumor Hospital lived in the same neighborhood (neighbors or relatives)" 2" Several controls at No" 6 Hospital were health care workers (nurses or doctors)" The numbers of neighbors or relatives (21, 52%) and health care workers (6, 50%) appeared to be higher than one might have expected if the selection of controls was random" All these "suspect" controls were outpatients with some very minor complaints, such as flu or upset stomach" An investigation of the 4 SHELL-MCCLURG-056310 selection procedures at these two hospitals was made in February and March, 2004. At the Tumor Hospital it was found that the Clinical Coordinator relied on one certain nurse for control selection. For some outpatient cases, the nurse "recruited" controls among her relatives or neighbors who fulfilled the gender and age matching criteria. Once an appropriate candidate had been identified, she asked the latter to register as an outpatient at the hospital (for a fee of 6), thus becoming eligible as a potential control who would be entitled to receive the 240 participation fee (equivalent to approximately a department store salesperson's salary for one week). At NO.6 Hospital, for some cases the Clinical Coordinator simply "recruited" coworkers (nurses and doctors) who met the matching criteria for gender and age. The same economic gain applied. The selection of controls from non-patients (relatives, neighbors and coworkers) will likely introduce selection bias. For co-workers, this procedure may possibly result in an under-representation of exposure among the controls. For relatives or neighbors, it is difficult to predict whether this selection procedure could bias study results. The primary motivation for this "unorthodox" approach appeared to be the 240 participation fee. In addition to the financial incentive, another motivation might be convenience. Two separate meetings with the Clinical Coordinators at the Tumor Hospital and NO.6 Hospital were held to discuss corrective measures. Given the different setups at each hospital and the number of individuals involved, the only "foolproof" method is to require all controls to be selected from inpatients. It is highly unlikely that a non-patient will be willing or able to register as an inpatient, if he or she is not actually sick. Of the 284 total cases in the study as of 20 April 2004, 26 (9%) are outpatients from the Tumor Hospital or NO.6 Hospital. Of these, we believe inappropriate controls were selected for 19 (73%) outpatient cases. Overall, inappropriate controls were selected for 7% of the cases (19/284). Our review did not detect any other irregularities in control selection, although we cannot guarantee that other inappropriate controls (possibly friends or relatives) were not selected for other cases. We do feel confident that this problem will be circumvented through the use of inpatient controls. While there may be some slight social class differences in outpatients and inpatients, we don't believe that these slight differences will cause a significant bias in the likelihood of occupational exposure. If such a bias is present, we believe it will be much smaller than the potential bias in selecting neighbors, co- 5 SHELL-MCCLURG-056311 workers, friends, or relatives. There is no ground to suspect that outpatients and inpatients differ substantially in any occupational or lifestyle factor. Dr. Ye Xibiao, the Project Coordinator at Fudan University, provided the following comparison between outpatients and inpatients. There are some minor differences in terms of smoking and age. Age is one of the matching variables, and, therefore, will not be an issue. Smoking will be treated as a confounding variable, and all analyses will be adjusted for smoking. 6 SH ELL-MCCLU RG-056312 Variables SEX Male Female Types of patients Outpatients Inpatients Total 78 60.5 51 39.5 401 55.6 320 44.4 479 56.4 371 43.6 MARTIAL STATUS DNK DIVORCED MARRIED NEVER MARRIED WIDOWED 0 0 111! 86.0) 12 9.3) 6( 4.7) 5( 0.7) 7( 1.0) 604( 83.8) 76( 10.5) 29( 4.0) 5( 0.6) 7( 0.8) 715( 84.1) 88( 10.4) 35( 4.1) EDUCATION DNK 1 0.8) HIGH MIDDLE 37 28.7) 35 27.1 NONE 5 3.9) POSTGRADUATE 3 2.3 PRIMARY UNIVERSITY 21 16.3) 27 20.9 3( 0.4) 196 27.2) 228 31.6 53 7.4) 4( 0.6) 111! 15.4) 126 17.5 4( 0.5) 233 27.4) 263 30.9 58 6.8) 7( 0.8) 132! 15.5) 153 18.0 SMOKE* No Yes, but not now Yes, still smoke 72 55.8 16 12.4 41 31.8 447 62.0 120 16.6 154 21.4 519 61.1 136 16.0 195 22.9 DRINK DNK No Yes 0 96! 74.4) 33 25.6 5( 0.7) 544! 75.5) 172 23.9 5( 0.6) 640! 75.3) 205 24.1 CITY Shanghai Out of Shanghai 99! 76.7) 30 23.3 586! 81.3) 685! 80.6) 135 18.7 16519.4 AGE(yrs)* 1835455565- 24 18.6 25 19.4 32 24.8 28 21.7 20 15.5 131( 18.2) 88( 12.2) 173 24.0 121 16.8 208 28.8 155 18.2 113 13.3 205 24.1 149 17.5 228 26.8 * p<O.05 7 SHELL-MCCLURG-056313