USING PROVINCIAL CLIENT REGISTRIES FOR SELECTION OF CONTROL SUBJECTS: LESSONS LEARNED

Volume 29-20
15 October 2003

[Table of Contents]

Case-control studies are frequently employed in epidemiologic research to examine postulated risk factor exposures in individuals with the disease of interest (cases) and those without the disease (controls). Developing a scientifically sound and cost-effective method for identifying and enrolling control subjects is a critical and challenging aspect of case-control study design. Control subjects should be selected independently of their exposure status to avoid and/or minimize bias as much as possible and ideally should be selected from the same source population as cases^(1,2). They can also be matched with cases on variables such as age, sex, and race, which might otherwise confound the relation between outcome and exposure variables^(3-5); however, care should be taken to avoid overmatching, which may partially or completely obscure evidence of a true causal association.

Historically, control subjects in case-control studies have been selected from a variety of sources, including hospitals, clinics, the neighbourhood, friends, or relatives^(1,6). They can be selected by systematic, convenience, or random sampling of the source population⁽¹⁾. These sources of control subjects and selection strategies are not without limitations. For example, responses of control subjects who are friends or relatives of a case could introduce bias because of their knowledge of the case's illness, and hospital- or clinic-based control subjects may be unrepresentative of the exposure distribution in the source population^(1,6).

Random digit dialing (RDD) and telephone directories may be used to select control subjects for telephone interviews. RDD frequently yields not-in-service or non-residential telephone numbers, and telephone directories may be out of date, have more than one telephone number per household, or not include unlisted telephone numbers. Both techniques usually entail numerous telephone calls in order to reach willing subjects in the desired age and sex groups^(1,6), and this may be particularly problematic in studies that require rapid enrolment of controls.

An alternative method of identifying matched healthy control subjects was successfully used in a recent outbreak investigation in British Columbia⁽⁷⁾. Control subjects were randomly selected from a provincial, population-based health care client registry.

This article outlines how client registries in British Columbia, Alberta, Saskatchewan, and Ontario were employed to generate age-matched control subjects for a 12-month, multi-provincial prospective case-control study of Salmonella Typhimurium infections. The goal of the article is to provide information to enable researchers to effectively use provincial client registry databases for selection of control subjects in future epidemiologic studies.

Materials and Methods

Basic health care in Canada is provided to most residents, except military personnel and the Royal Canadian Mounted Police, through the provincial/territorial ministries of health (a mandatory requirement of the Canada Health Act⁽⁸⁾). Provincial/territorial client registry listings include the names, date of birth, sex, address, and telephone numbers for each insured resident and are periodically updated. There are some differences among provinces/territories - for example, in Ontario, newborns whose parents hold a valid health number and have completed an infant registration form are given a health number, but in British Columbia, until recently, newborns were billed on their mother's health number for the first 3 months.

The methods for the multi-provincial Salmonella Typhimurium case-control study were jointly developed by the participating provinces and Health Canada, Foodborne, Waterborne and Zoonotic Infections Division. Rigorous efforts were made to maintain scientific comparability among provinces, although each province made minor modifications to the protocol for administrative purposes. Ethical approval for this study was obtained from the University of Guelph, Ontario, the University of Regina, Saskatchewan, and the University of British Columbia, Vancouver, British Columbia.

The study, which aimed to identify risk factors and the burden of illness associated with S. Typhimurium infection, consisted of telephone interviews of laboratory-confirmed cases and age-matched control subjects identified during the 1-year study period (1 December, 1999, to 30 November, 2000). An eligible case was defined as an individual residing in the study province who experienced a diarrheal illness during the study period and from whose stool sample S. Typhimurium was isolated by the provincial public health laboratory. For each case, an age-matched control subject was randomly selected from the provincial client registry.

Review of provincial S. Typhimurium case data was undertaken to determine the expected number and age-band distribution of cases^(9,10). A stratified random sample of control subjects was obtained - 10 times the expected number of cases, in order to allow for wrong numbers and nonparticipation - by the corresponding participating provincial public health authority.

Control subjects were matched with cases by age band (Table 1) in all provinces except one, where the database format required that they be matched to cases by exact date of birth. Age band 1, which consisted of newborns 0 to 5 months of age, was re-sampled in two provinces to replace infants who had left this age category during the study. Newborn control subjects who had not previously been included in the study (“unused”) were promoted to age band 2 upon reaching 6 months of age, and the list was re-randomized.

Two provinces sent information letters to potential control subjects (or parents/guardians in the case of children) before the study began and again when the number of unused control subjects in any age group became too few to meet the requirements of the study protocol. Control subjects were excluded if they were unable to communicate in English, could not be reached by telephone after six attempts (over 2 days, including evenings), or if they reported having diarrhea in the 28 days before the interview. If the first control subject was ineligible, a second was obtained from the database and six attempts were again made to contact the individual. This procedure was repeated until either a control subject had been reached or 7 days had passed since the case interview, and then the case was removed from the study.

Trained interviewers conducted all interviews in each province using scripted questionnaires and an interviewer's manual to facilitate consistent data collection. Introductory scripts specifically created for three age groups (< 12 years of age, 12 to 17, and >= 18) provided a description of the study and the anticipated interview length (25 minutes), and requested consent to participate.

Interviewers documented the outcome of each call attempt on a standardized call record form. Information collected on the call record form included telephone number, date, call outcome (i.e. busy, no answer, answering machine, subject not at home, and misconnects*) and willingness to participate. These data were entered into Microsoft Excel 2000, categorized by province, and analyzed descriptively using Microsoft Excel 2000 and SAS version 6. Saskatchewan was not included in the analysis because of small numbers; the remaining three provinces (British Columbia, Alberta, and Ontario) were randomly denoted Province I, II, and III.

Table 1. Participation rates and number of telephone numbers required to recruit healthy controls, by age band, all provinces
Age band	Age range	Participants	Non-participants	Participation rate, %	Telephone numbers used	Telephone numbers used per participant
1	0-5 mo	2	3	40.0	21	10.5
2	6-11 mo	13	1	92.9	32	2.5
3	12-23 mo	23	16	59.0	103	4.5
4	2-4 yrs	64	11	85.3	281	4.4
5	5-9 yrs	55	14	79.7	231	4.2
6	10-19 yrs	47	17	73.4	251	5.3
7	20-29 yrs	42	32	56.8	221	5.3
8	30-39 yrs	46	44	51.1	283	6.2
9	40-49 yrs	24	11	68.6	112	4.7
10	50-59 yrs	22	21	51.2	112	5.1
11	60-69 yrs	15	12	55.6	59	3.9
12	> 70 yrs	12	7	63.2	36	3.0
Unknown		13	5	72.2	46	3.5
Overall		378	194	66.1	1,788	4.7

Results

Participation Rates

Participation rates overall and for each province are presented in Tables 1 and 2. A total of 572 clients were reached and, of these, 378 (66%) agreed to participate. Participation rates varied by age band: overall rates for age bands 1, 8 and 10 were low (< 52%), whereas in age bands 2 and 4 they were higher (> 85%). Province I had the highest rate of participation, 83.4%, followed by province III, at 59.2%, and province II, at 51.9%. The mean number of telephone numbers used and total telephone calls made to contact a willing participant also varied among provinces, provinces I, II, and III requiring 3.8, 6.9, and 3.3 telephone numbers and 6.8, 10.2, and 5.9 telephone calls respectively.

Age bands 2 and 4 showed relatively high participation rates in all three provinces, but participation in the other age bands tended to be lower overall.

Proportion of In-service Residential Telephone Numbers

In all provinces, for the 1788 telephone numbers tried, 2813 telephone calls (1.6 calls/number) were made. Of the 1788 numbers, 1433 (80.1%) were in-service residential numbers, and the remainder (355, 19.8%) were misconnects. Overall, age bands 2, 11, and 12 displayed high proportions of in-service residential telephone numbers, of 90.5%, 96.6%, and 88.8% respectively. Province I had the highest proportion of in-service residential numbers (85%) followed by province II (79%) and province III (68%). In province I, there was little variation among the age bands in this respect: only one small group (age band 1) had < 79% in-service residential telephone numbers. In province II, the higher proportion of in-service residential numbers occurred in age bands representing those < 2 years of age (age bands 1, 2, and 3) and > 60 (age bands 11 and 12). Province III showed greater variation in the proportion of these telephone numbers among age bands than the other two provinces. As in province II, those in the lower age bands (< 4 years of age, age bands 2, 3, and 4) and the higher ones ( >= 60, age bands 10 to 12) accounted for a higher proportion of in-service residential telephone numbers than those falling into the middle age band categories.

Discussion

Using provincial client registry listings as a source of population-based control subjects provides a number of advantages that other sources may not offer. First, the information in the registry may be updated after medical encounters, and this improves the efficiency of reaching potential control subjects. Second, since individual-level demographic information such as sex, age, and geographic location is available from the registry, matching control subjects with cases on these variables is a relatively quick and straightforward process. Finally, use of client registries, which are electronic data sets, makes random selection of control subjects from the same population (study base) as the cases technically simple⁽⁵⁾.

The overall participation rate (defined as the total number agreeing to participate/total number agreeing to participate plus the number of refusals) in our study (66%) compares favourably with published participation rates of 38% to 80.1% when RDD is used, even though our participants were informed of the expected duration of the interview^(11-13). Recruitment/information letters designed for different age groups were sent out in some provinces before the study began in order to inform potential control subjects of its purpose. These letters likely had a positive effect on response rate, in that they prepared recipients for the possibility of a future telephone interview and enhanced the credibility of the study.

The results of the study suggest that those in the younger and older age bands are more likely to have correct information in the client database. This may be a result of frequent or more recent medical visits by individuals in these age bands resulting in updating of the registry. Province I had the highest proportion of in-service residential telephone numbers overall and in most age categories as compared with the other provinces. This may reflect the varying frequency and speed at which the database is updated after medical encounters among the different provinces.

The distribution of in-service residential telephone numbers varied among provinces and within each province depending upon the age band. Only one age band in one province (one participant) required more than nine telephone numbers to enroll the required participant. For future studies using client registries for the selection of control subjects, it is recommended that a sample 10 times the number needed in each age group be obtained to ensure that there are sufficient telephone numbers to enroll the required number of participants. In this study, an average of 7.5 telephone calls per participant was required; this information will assist researchers in estimating interviewers' time and monetary budgets. Some provinces charge for use of the client registry information, so this should also be included in budget considerations. Unfortunately, the sex of clients contacted and participating control subjects was not recorded on the data collection sheet; therefore it was not possible to assess whether there was a difference in response rate by sex.

There are limitations associated with the use of client registries for selection of control subjects. Registries may not include all residents in a province: military and RCMP personnel, individuals resident in the province for < 3 months or, as previously mentioned, newborns in some provinces may be excluded. Another limitation identified by some of the provinces was the delay involved in obtaining legal approval to use the registries. It may be worthwhile in future to overcome the legalities involved before beginning the study; awareness of this issue will help to plan the timing of the study.

In conclusion, using client registries to select control subjects is a valid and feasible technique that should be considered in future case-control studies in Canada.

Table 2. Participation rates, number and validity of telephone numbers, and calls required to recruit healthy controls, by age band and province
	Age band	Participants	Non-participants	Participation rate, %	Telephone numbers used
PROVINCE I	1	1	1	50.0	6
	2	9	0	100.0	25
	3	13	3	81.3	35
	4	43	6	87.8	196
	5	30	10	75.0	130
	6	20	1	95.2	74
	7	16	3	84.2	59
	8	15	2	88.2	66
	9	16	3	84.2	55
	10	12	5	70.6	46
	11	8	4	66.7	27
	12	8	1	88.9	14
	Unknown	5	0	100.0	8
	Total	196	39	83.4	741
PROVINCE II	1	0	1	0.0	7
	2	1	1	50.0	2
	3	8	8	50.0	57
	4	18	5	78.3	80
	5	12	2	85.7	72
	6	21	11	65.6	151
	7	20	23	46.5	138
	8	22	36	37.9	194
	9	6	4	60.0	45
	10	6	12	33.3	51
	11	4	5	44.4	25
	12	1	4	20.0	13
	Unknown	5	3	62.5	21
	Total	124	115	51.9	856
PROVINCE III	1	1	1	50.0	8
	2	3	0	100.0	5
	3	2	5	28.6	11
	4	3	0	100.0	5
	5	13	2	86.7	29
	6	6	5	54.5	26
	7	6	6	50.0	24
	8	9	6	60.0	23
	9	2	4	33.3	12
	10	4	4	50.0	15
	11	3	3	50.0	7
	12	3	2	60.0	9
	Unknown	3	2	60.0	17
	Total	58	40	59.2	191
* Misconnects includes wrong numbers, numbers not in service, fax, business and cell phone numbers. ^† In service residential phone number refers to those numbers successfully used to reach the person intended.

Table 2. (continued) Participation rates, number and validity of telephone numbers, and calls required to recruit healthy controls, by age band and province
	Age band	Telephone numbers used per participant	No. of phone calls needed to contact a participant	Total number of misconnects*	Proportion of in-service residential phone numbers^†
PROVINCE I	1	6.0	9	2	0.67
	2	2.8	38	3	0.88
	3	2.7	64	5	0.86
	4	4.6	271	41	0.79
	5	4.3	234	22	0.83
	6	3.7	102	10	0.86
	7	3.7	77	11	0.81
	8	4.4	102	9	0.86
	9	3.4	102	2	0.96
	10	3.8	84	5	0.89
	11	3.4	38	1	0.96
	12	1.8	20	0	1.00
	Unknown	1.6	16	1	0.88
	Total	3.8	1,157	112	0.85
PROVINCE II	1	0.0	9	1	0.86
	2	2.0	2	0	1.00
	3	7.1	63	9	0.84
	4	4.4	95	19	0.76
	5	6.0	92	16	0.78
	6	7.2	201	32	0.79
	7	6.9	259	30	0.78
	8	8.8	298	50	0.74
	9	7.5	66	8	0.82
	10	8.5	97	10	0.80
	11	6.3	30	0	1.00
	12	13.0	21	1	0.92
	Unknown	4.2	26	5	0.76
	Total	6.9	1,259	181	0.79
PROVINCE III	1	8.0	18	3	0.63
	2	1.7	15	0	1.00
	3	5.5	16	2	0.82
	4	1.7	14	1	0.80
	5	2.2	51	9	0.69
	6	4.3	54	12	0.54
	7	4.0	55	9	0.63
	8	2.6	41	7	0.70
	9	6.0	35	4	0.67
	10	3.8	35	2	0.87
	11	2.3	12	1	0.86
	12	3.0	16	3	0.67
	Unknown	5.7	35	9	0.47
	Total	3.3	397	62	0.68
* Misconnects includes wrong numbers, numbers not in service, fax, business and cell phone numbers. ^† In service residential phone number refers to those numbers successfully used to reach the person intended.

Acknowledgement

The authors wish to thank the interviewers (in particular Michelle Cox and Gary Svoboda) for their assistance with the data collection phase of this study.

References

Rothman KJ, Greenland S. Case-control studies. In: Modern epidemiology. Lippincott: Williams and Wilkins, 1998.
Wacholder S, McLaughlin JK, Silverman DT et al. Selection of controls in case-control studies: principles. Am J Epidemiol 1992;135(9):1019-28.
Kelsey JL, Thompson WD, Evans AS. Case-control studies. In: Methods in observational epidemiology. New York: Oxford University Press, 1986.
Wacholder S, McLaughlin JK, Silverman DT et al. Selection of controls in case-control studies: design options. Am J Epidemiol 1992;135(9):1042-50.
Breslow N. Design and analysis of case-control studies. Annu Rev Public Health 1982;(3):29-54.
Wacholder S, McLaughlin JK, Silverman DT et al. Selection of controls in case-control studies: types of controls. Am J Epidemiol 1992;135(9):1029-41.
Nowgesic E, Fyfe M, Hockin J et al. Outbreak of Yersinia pseudotuberculosis in British Columbia - November 1998. CCDR 1999;25(11):97-100.
8. Health Canada. Canada Health Act annual report 2000-2001. Chapter p.1-7:133-43.
Middleton D, Ciebin B, Michel P et al. Salmonella Typhimurium definitive type 104 in Ontario, 1997-1998: one example of an antimicrobial resistant organism. Public Health Epidemiol Rep Ont 1999;10:230-35.
Buxton J, Fyfe M, King A et al. Salmonella Typhimurium definitive type 104 isolates in British Columbia, 1997-98. CCDR 1999;25(15):129-33.
Patten BS. Major depression prevalence in Calgary. Can J Psychiatry 2000;45:923-26.
Leech JA, Wilby K, McMullen E et al. The Canadian Human Activity Pattern Survey: report of methods and population surveyed. Chron Dis Can 1996;17:118-123.
A survey of Toronto residents awareness: uses and attitudes towards lawn pesticides. Toronto: Toronto Public Health, Health Promotion and Environmental Protection Office, April 2002. URL: <www.city.toronto.on.ca/health/>.

Source: B Krenten-Boaretto, BSc, Department of Population Medicine, Ontario Veterinary College, University of Guelph; JA Buxton, MBBS, MHSc , Department of Health Care and Epidemiology, University of British Columbia, Vancouver; K Doré, MHSc, Foodborne, Waterborne and Zoonotic Infections Division, Centre for Infectious Disease Prevention and Control, Public Health Agency of Canada, Health Canada, Guelph; M Fyfe, MD, MSc, British Columbia Centre for Disease Control, Vancouver; D Middleton, DVM, MSc, Ontario Ministry of Health and Long-Term Care, Disease Control Service, Public Health Branch, Toronto; S McEwen, DVM, DVSc, Department of Population Medicine, Ontario Veterinary College, University of Guelph; and the Multi-Provincial Salmonella Typhimurium Case-Control Study Group:

M Fyfe, British Columbia Centre for Disease Control; J Buxton, Department of Health Care and Epidemiology, University of British Columbia; A King, Immunization and Respiratory Infections Division, Health Canada; A Paccagnella, British Columbia Centre for Disease Control; K Grimsrud, Alberta Health & Wellness; I Zazulak, Capital Health, Edmonton; J Talbot and R Rennie, Provincial Laboratory of Public Health for Northern Alberta; P Pieroni, Laboratory & Disease Control Services Branch, Saskatchewan Health; R Ahmed and F Rodgers, National Laboratory for Enteric Pathogens, Health Canada; F Pollari, K Doré and J Wilson, Foodborne, Waterborne and Zoonotic Infections Division, Health Canada; P Michel, Laboratory for Foodborne Zoonoses, Health Canada; D Middleton, M Naus, B Henry, B Cieben, and F Jamieson, Ontario Ministry of Health and Long-Term Care.

[Table of Contents]

* Misconnects includes wrong numbers, numbers not in service, fax, business and cell phone numbers.


	Last Updated: 2003-10-15		Important Notices