Abstract
Random-digit dialing (RDD) using landline telephone numbers is the historical gold standard for control recruitment in population-based epidemiologic research. However, increasing cell-phone usage and diminishing response rates suggest that the effectiveness of RDD in recruiting a random sample of the general population, particularly for younger target populations, is decreasing. In this study, we compared landline RDD with alternative methods of control recruitment, including RDD using cell-phone numbers and address-based sampling (ABS), to recruit primarily white men aged 18–55 years into a study of testicular cancer susceptibility conducted in the Philadelphia, Pennsylvania, metropolitan area between 2009 and 2012. With few exceptions, eligible and enrolled controls recruited by means of RDD and ABS were similar with regard to characteristics for which data were collected on the screening survey. While we find ABS to be a comparably effective method of recruiting young males compared with landline RDD, we acknowledge the potential impact that selection bias may have had on our results because of poor overall response rates, which ranged from 11.4% for landline RDD to 1.7% for ABS.
Keywords: case-control studies, data collection, epidemiologic methods, postal service, telephone, testicular neoplasms
Random-digit dialing (RDD) is the historical gold standard for population-based control recruitment in epidemiologic research. Traditionally, RDD sampling frames have included only landline telephone numbers. In the 1990s, RDD was shown to be a reliable method for selecting an equal probability sample of landline numbers in areas where telephone ownership was high (1). However, in the past decade, researchers have reported a decline in response rates for RDD screening. In a report on trends in RDD control recruitment for childhood cancer studies, Bunin et al. (2) found that the response rate decreased 2.5% per year over a 20-year period. Other studies have suggested that an increase in cell-phone usage, among other factors, may explain this trend (3). A 2012 national survey of households found that a majority (60.1%) of persons aged 25–29 years used a cell phone exclusively, while one-tenth (10.5%) of adults aged 65 years or older did (4). Consequently, RDD may not be the most effective method of recruiting certain target populations, particularly younger persons, who are less likely to own a landline phone.
Researchers have employed other methods to include cell-phone-only households in samples of the general population (5). Cell-phone RDD involves using telephone exchanges that correspond to cell-phone numbers as opposed to landline numbers. In a recent case-control study, Voigt et al. (3) demonstrated the feasibility of cell-phone RDD but found that this method had a screening rate nearly half that of landline RDD. Furthermore, a survey research study found that cell-phone RDD may result in the identification of a population with different behaviors and characteristics than populations identified through other methods, such as address-based sampling (ABS) (6).
ABS is an alternative sampling method that involves mailing screening surveys to addresses that have been selected at random from an electronic database. Researchers have suggested that the Delivery Sequence File used by the US Postal Service may allow for the selection of sampling frames that surpass the coverage of those used for landline RDD (7, 8). One study comparing the efficacy of RDD and ABS methods in reaching respondents for a multistate health survey found that ABS was more cost-effective than RDD, allowed for access to cell-phone-only households, and resulted in higher response rates in some US states; however, it overrepresented non-Hispanic whites and persons who were better educated (9). In contrast, in a smaller survey research study, Johnson et al. (10) reported similar response rates when comparing RDD with ABS and noted no differential representation by Hispanic ethnicity.
To our knowledge, there has been no published report of a comparison between RDD and ABS methods of control recruitment in a case-control study. Therefore, we evaluated the use of landline RDD, cell-phone RDD, and ABS methods for recruiting young men in an ongoing case-control study of testicular cancer susceptibility at the University of Pennsylvania. We compared participation and respondent characteristics among those recruited to evaluate the use of RDD and ABS as recruitment methods for future population-based research, particularly that targeting younger populations.
MATERIALS AND METHODS
We sought to identify a comparison group of 1,500 eligible male control subjects frequency-matched by age, race, and geographical location to enrolled case subjects with testicular cancer. Landline and cell-phone RDD methods were initially used to recruit the sample. Of this sample, it was estimated that 71% of the persons contacted would complete all study requirements, yielding 1,100 enrolled controls, including 1,000 white male control subjects. Because of low yields of eligible controls from these initial screenings, an ABS method was subsequently used to complete control recruitment. The study protocol was approved by the Institutional Review Board of the University of Pennsylvania.
RDD recruitment and data collection
An outside research organization (Abt SRBI, Silver Spring, Maryland) was contracted to recruit controls via telephone from 8 counties in the Philadelphia, Pennsylvania, area from which cases for the testicular cancer study had been previously identified, including: Bucks County, Pennsylvania; Burlington County, New Jersey; Camden County, New Jersey; Chester County, Pennsylvania; Delaware County, Pennsylvania; Gloucester County, New Jersey; Montgomery County, Pennsylvania; and Philadelphia County, Pennsylvania. Potential controls were contacted in 2 waves between January and February 2009 and April and May 2009. For both waves, the landline sample was drawn using a systematic random sample from a list-assisted RDD frame including all 100 blocks with at least 1 listed phone number in the targeted counties (11). During the second wave, potential controls were also contacted by cell phone. The cell-phone RDD sample was drawn using a systematic random sample stratified by county and service provider from a frame of 100 blocks built from activated wireless phone numbers. The landline and wireless frames were mutually exclusive.
Persons who were contacted via landline or cell phone and agreed to answer questions were screened according to gender, age, race/ethnicity, and history of testicular cancer. Telephone numbers were dialed a total of 10 times to find a qualified individual in the household. Persons contacted by cell phone were asked questions to assess the safety of cell-phone use at the time of the call (i.e., that the person was not driving while answering the call). Contactees were first screened on the basis of age and gender inclusion criteria, and only males between the ages of 18 and 52 years were considered for further screening. Quotas for specific age groups (within the range 18–52 years) and race/ethnicity were used to frequency-match controls to the age and race/ethnicity distribution of enrolled cases. Persons who passed these initial screenings were administered the full screening survey.
The full screening survey contained questions on general health status (excellent/very good/good/fair/poor), utilization of primary care (every 6 months or more often/once a year/once every 2 years/less than every 2 years/only when sick/never), paternity (yes/no), number of siblings, number of older siblings, and prior history of cancer (yes/no), as well as a question on history of testicular cancer (yes/no), which was used to screen out affected persons. Questions on sibship were asked, in part, to identify whether the respondent was the firstborn among his siblings. Evidence exists that the prenatal environment may differ between firstborns and those born after a previous full-term pregnancy (12), and we ultimately sought to examine whether prenatal environment affects susceptibility to testicular cancer later in life. Questions about general health status, number of siblings, and history of cancer other than testicular cancer were only asked during the second wave of screening. On average, the full recruitment interview in each wave took approximately 7 minutes to complete. Eligible respondents who completed the full screening survey and agreed to be recruited for the study were sent the research study questionnaire and consent form. The study questionnaire included questions on phone usage and behavior. When a completed study questionnaire and consent form were returned, participants were enrolled in the study and sent a saliva collection kit so they could provide a specimen for genetic analysis. If the study questionnaire and consent form were not returned within 2 weeks, respondents were sent weekly reminder letters for 3 weeks or until the study questionnaire and consent form were returned. Those men who returned a specimen were sent a $25 gift certificate redeemable at a national electronics retail chain.
ABS recruitment and data collection
The US Postal Service's Delivery Sequence File contains a listing of all addresses to which mail is delivered in the United States. The Delivery Sequence File list of addresses was obtained for 12 counties in the Philadelphia area from which cases for the testicular cancer study were selected from 2010 to 2012, including: Berks County, Pennsylvania; Bucks County, Pennsylvania; Burlington County, New Jersey; Camden County, New Jersey; Chester County, Pennsylvania; Delaware County, Pennsylvania; Gloucester County, New Jersey; Lancaster County, Pennsylvania; Lehigh County, Pennsylvania; Montgomery County, Pennsylvania; Northampton County, Pennsylvania; and Philadelphia County, Pennsylvania (geographical inclusion criteria were extended to 4 additional counties between the RDD and ABS phases of the study). ABS recruitment mailings took place during the month of March 2010 (pilot phase) as well as between January 2011 and June 2012. The ABS frame excluded business addresses and consisted of residential city-style and post office box addresses. During the pilot phase, seasonal, educational, vacant, throwback, and drop-unit addresses were also excluded. A random sampling of addresses, frequency-matched to the census tract distribution of men with testicular cancer (matching was not undertaken in the pilot phase), was selected to recruit controls for the study.
Selected addresses were sent an introductory letter that explained the project and the target population (men aged 18–55 years) and a screening survey, which was based on that used in the second wave of the RDD screening. Respondents who returned a completed screening survey and were deemed eligible for the study were sent the research study questionnaire and a consent form. Respondents were deemed eligible if they were male, had no history of testicular cancer, and were between the ages of 18 and 55 years (to increase the number of eligible men, the upper limit of the age eligibility criterion was extended from 52 to 55 years between the RDD and ABS phases of the study). If the study questionnaire and consent form were not returned within 2 weeks, respondents were sent weekly reminder letters for 3 weeks or until the study questionnaire and consent form were returned. Similar to the RDD group, if a completed questionnaire and consent form were returned, participants were enrolled in the study and sent a saliva collection kit so they could provide a specimen for genetic analysis. Those men who returned a specimen were sent a $25 gift certificate redeemable at a national electronics retail chain.
Statistical analysis
Response outcomes for both RDD and ABS screening were categorized using the method of Olson et al. (13). For RDD, a phone number was considered an ineligible sampling unit if a business, government, nonresident, data-line, or nonworking number was reached, or if there was an error when dialing or trouble with the telephone connection. In addition, a cell-phone number was considered an ineligible sampling unit in the landline RDD screening. All other dialed phone numbers were considered eligible sampling units.
Based on response outcome categories, response rates for RDD screening were calculated using the method of Slattery et al. (14). For ABS screening, the overall response rate was defined as the percentage of screening surveys received among all of the surveys sent out. The field response rate was defined as the percentage of respondents who completed a study questionnaire and consent form among all screened respondents who were deemed eligible to participate in the study.
Responses to questions from the screening survey and study questionnaire are reported as a percentage of total responses, after excluding nonresponses, which were rare. For questions about phone usage, responses from controls were compared with those of men with testicular cancer (cases), who answered a questionnaire similar to that administered to controls. For comparisons of characteristics between groups, chi-squared, Fisher's exact, and Spearman's rank correlation tests were used where appropriate, and corresponding odds ratios and 95% confidence intervals are presented. To account for potential confounding related to the expected difference in the distribution of age, race, and geography that arose because of the implementation of different sampling procedures in RDD and ABS recruitment, we used logistic regression modeling to adjust for these covariates; corresponding adjusted odds ratios and 95% confidence intervals are presented. Responses of ABS- and RDD-recruited controls to the question regarding health status were compared with responses to the same question from the 2010 National Health Interview Survey, with data restricted to males aged 18–55 years living in the Northeast region of the United States (15). All analyses were performed using Stata, version 12.1 (StataCorp LP, College Station, Texas).
RESULTS
Response outcomes and rates
RDD screening
A total of 18,361 eligible sampling units were dialed using landline RDD screening (Table 1). A total of 5,603 persons reached by telephone (13.8%) were screened out as ineligible because of female gender, an age outside of the 18- to 52-year inclusion criterion, or a history of testicular cancer. Additionally, age and race/ethnicity quotas were used to further screen out participants in order to recruit a control population that was frequency-matched to the case population with regard to these characteristics. In total, 323 eligible persons (0.8%) were identified and asked to complete the full screening survey. Of this group, 119 (36.8%) provided study consent, most of whom (35.3%) completed the study requirements, including return of a saliva sample. The screening contact, screening cooperation, and screening response rates were 59.6%, 54.2%, and 32.3%, respectively (Table 2). The field response rate was 35.3%; among men aged 18–39 years, it was 32.7%. The overall response rate was 11.4%.
Table 1.
Response Outcomes for the Random-digit Dialing and Address-based Sampling Screening Methods, Philadelphia, Pennsylvania, 2009–2012
| RDD |
ABS |
|||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Landlines |
Cell Phones |
|||||||||||
| Total No. | No. Excluded | % of Total | % Excluded | Total No. | No. Excluded | % of Total | % Excluded | Total No. | No. Excluded | % of Total | % Excluded | |
| Ineligible sampling unit | 22,238 | 54.8 | 1,469 | 36.7 | ||||||||
| Business, government, nonresidential | 2,549 | 11.5 | 76 | 5.2 | ||||||||
| Cell phone | 14 | 0.1 | ||||||||||
| Data line | 2,002 | 9.0 | 4 | 0.3 | ||||||||
| Nonworking number | 16,123 | 72.5 | 1,300 | 88.5 | ||||||||
| Other | 1,550 | 7.0 | 89 | 6.1 | ||||||||
| Unable to determine eligibility | 12,435 | 30.6 | 1,498 | 37.5 | 65,597 | 96.8 | ||||||
| Unknown whether residential | 5,334 | 42.9 | 198 | 13.2 | ||||||||
| Residential, unknown whether person was eligible | 7,101 | 57.1 | 1,300 | 86.8 | ||||||||
| Answering machine/voice mail | 1,974 | 15.9 | 675 | 45.1 | ||||||||
| Non-English-speaking respondent | 117 | 0.9 | 30 | 2.0 | ||||||||
| Respondent refused to answer questions on eligibility | 4,007 | 32.2 | 586 | 39.1 | ||||||||
| Other | 1,003 | 8.1 | 9 | 0.6 | ||||||||
| Questions on eligibility left blank | 15 | <0.1 | ||||||||||
| No response to letter | 65,582 | >99.9 | ||||||||||
| Respondent not eligiblea | 5,603 | 13.8 | 1,013 | 25.3 | 332 | 0.5 | ||||||
| Gender/age | 4,998 | 89.2 | 906 | 89.4 | 314 | 94.6a | ||||||
| History of testicular cancer | 4 | 0.1 | 22 | 6.6a | ||||||||
| Age quota already met | 498 | 8.9 | 83 | 8.2 | ||||||||
| Race/ethnicity quota already met | 103 | 1.8 | 24 | 2.4 | ||||||||
| County | 8 | 2.4a | ||||||||||
| Respondent screened and eligible | 323 | 0.8 | 20 | 0.5 | 1,821 | 2.7 | ||||||
| Agreed to further contact | 221 | 68.4 | 9 | 45.0 | ||||||||
| Consent provided | 119 | 36.8 | 3 | 15.0 | 1,021 | 56.1 | ||||||
| Completeb | 114 | 35.3 | 2 | 10.0 | 969 | 53.2 | ||||||
| Not complete | 5 | 1.5 | 1 | 5.0 | 52 | 2.9 | ||||||
| Consent not provided | 102 | 31.6 | 6 | 30.0 | 800 | 43.9 | ||||||
| Refused further contact | 102 | 31.6 | 12 | 60.0 | ||||||||
| Total | 40,599 | 100 | 4,000 | 100 | 67,750 | 100 | ||||||
Abbreviations: ABS, address-based sampling; RDD, random-digit dialing.
a Percentages may not total 100% because respondents could satisfy multiple exclusion criteria.
b Completed all required study components, including signed consent form, questionnaire, and biospecimen.
Table 2.
Response Rates for Landline Random-digit Dialing, Cell-Phone Random-digit Dialing, and Address-based Sampling, Philadelphia, Pennsylvania, 2009–2012
| RDD, % |
ABS, % | ||
|---|---|---|---|
| Landlines | Cell Phones | ||
| Screening contact ratea | 59.6 | 64.3 | |
| Screening cooperation rateb | 54.2 | 63.5 | |
| Screening response ratec | 32.3 | 40.8 | 3.2 |
| Field response rated | 35.3 | 10.0 | 53.2 |
| Field response rate, ages 18–39 yearse | 32.7 | 10.0 | 51.8 |
| Overall response ratef | 11.4 | 4.1 | 1.7 |
Abbreviations: ABS, address-based sampling; RDD, random-digit dialing.
a Percentage of household units with which there was contact (a person answered the phone) among all eligible sampling units.
b Percentage of participating households (a person agreed to answer at least some screening questions) among all household units with which there was contact.
c For RDD, defined as the percentage of participating households among all eligible sampling units; for ABS, defined as the percentage of screening surveys received among all surveys sent out.
d For RDD, defined as the percentage of respondents who completed a study questionnaire and consent form among all screened respondents who were deemed eligible and agreed to further contact; for ABS, defined as the percentage of respondents who completed a study questionnaire and consent form among all screened respondents who were deemed eligible.
e For RDD, defined as the percentage of respondents aged 18–39 years who completed a study questionnaire and consent form among all screened respondents aged 18–39 years who were deemed eligible; for ABS, defined as the percentage of respondents aged 18–39 years who completed a study questionnaire and consent form among all screened respondents aged 18–39 years who were deemed eligible.
f Product of the screening response rate and the field response rate.
Response outcomes among those contacted during cell-phone RDD screening are also given in Table 1. The screening contact, screening cooperation, and screening response rates of 64.3%, 63.5%, and 40.8%, respectively, were higher than those for landline RDD screening (Table 2). Conversely, the field response rate (10.0%) and the rate among men aged 18–39 years (10.0%) were lower than those for landline RDD recruitment, as was the overall response rate (4.1%).
ABS screening
A total of 67,750 screening surveys were mailed out during the ABS screening phase of the study (Table 1). A total of 2,153 of these surveys (3.2%) were returned by male respondents, and 332 (0.5%) were deemed ineligible. Of the 1,821 eligible respondents (2.7%), 1,021 (56.1%) completed a consent form, of whom 969 (53.2%) also completed all study requirements (i.e., the field response rate). The field response rate among men aged 18–39 years was 51.8%. The overall response rate was 1.7% (Table 2).
Respondent characteristics
Characteristics of eligible controls (all respondents who completed the screening survey and were deemed eligible for the study) and enrolled controls (all eligible respondents who completed a consent form and questionnaire) are given in Tables 3 and 4, respectively. After adjusting for age, race, and county, controls recruited by ABS were similar to those recruited by RDD with regard to primary-care utilization, likelihood of having siblings, number of older siblings, and history of cancer other than testicular cancer. Eligible controls recruited by RDD were more likely to report fair or poor heath (adjusted odds ratio (AOR) = 2.27, 95% confidence interval (CI): 1.15, 4.46) than those recruited by ABS, which was largely attributable to men aged 18–29 years (after adjustment for race and county, AOR = 8.06, 95% CI: 2.22, 29.24). Both eligible and enrolled controls recruited by RDD were more likely to have fathered a child than those recruited by ABS (AOR = 1.56 (95% CI: 1.15, 2.12) and AOR = 2.43 (95% CI: 1.43, 4.12), respectively).
Table 3.
Characteristics of Eligible Controls Recruited by Address-based Sampling Compared With Those Recruited by Random-digit Dialing, Philadelphia, Pennsylvania, 2009–2012
| Characteristic | ABS |
RDD |
ORa | 95% CI | AORb | 95% CI | ||
|---|---|---|---|---|---|---|---|---|
| No. | % | No. | % | |||||
| Age group, years | ||||||||
| 18–29 | 324 | 17.8 | 117 | 34.1 | 1.00 | Referent | ||
| 30–39 | 446 | 24.5 | 123 | 35.9 | 0.77 | 0.57, 1.03 | ||
| 40–49 | 569 | 31.2 | 77 | 22.4 | 0.38 | 0.27, 0.52 | ||
| 50–55 | 482 | 26.5 | 26 | 7.6 | 0.15 | 0.10, 0.23 | ||
| Total | 1,821 | 343 | ||||||
| Race | ||||||||
| Nonwhite | 245 | 13.5 | 31 | 9.0 | 1.00 | Referent | ||
| White | 1,576 | 86.5 | 312 | 91.0 | 1.56 | 1.06, 2.32 | ||
| Total | 1,821 | 343 | ||||||
| Geographical areac | ||||||||
| Region 1 | 297 | 16.3 | 56 | 16.3 | 1.00 | Referent | ||
| Region 2 | 720 | 39.5 | 186 | 54.2 | 1.37 | 0.99, 1.90 | ||
| Region 3 | 522 | 28.7 | ||||||
| Region 4 | 282 | 15.5 | 101 | 29.4 | 1.90 | 1.32, 2.74 | ||
| Total | 1,821 | 343 | ||||||
| Health status | ||||||||
| Excellent | 462 | 25.4 | 46 | 26.9 | 1.00 | Referent | 1.00 | Referent |
| Very good | 789 | 43.4 | 69 | 40.4 | 0.88 | 0.59, 1.30 | 0.98 | 0.65, 1.48 |
| Good | 450 | 24.7 | 41 | 24.0 | 0.92 | 0.59, 1.42 | 1.35 | 1.85, 2.16 |
| Fair/poor | 119 | 6.5 | 15 | 8.8 | 1.27 | 0.68, 2.35 | 2.27 | 1.15, 4.46 |
| Total | 1,820 | 171 | ||||||
| Primary-care utilizationd | ||||||||
| Regular | 1,082 | 59.5 | 202 | 58.9 | 1.00 | Referent | 1.00 | Referent |
| Irregular | 641 | 35.2 | 130 | 37.9 | 1.09 | 0.85, 1.38 | 1.03 | 0.79, 1.34 |
| Never | 97 | 5.3 | 11 | 3.2 | 0.61 | 0.32, 1.15 | 0.52 | 0.26, 1.04 |
| Total | 1,820 | 343 | ||||||
| Ever having fathered a child | ||||||||
| No | 644 | 35.4 | 140 | 40.8 | 1.00 | Referent | 1.00 | Referent |
| Yes | 1,176 | 64.6 | 203 | 59.2 | 0.79 | 0.63, 1.01 | 1.56 | 1.15, 2.12 |
| Total | 1,820 | 343 | ||||||
| Having siblings | ||||||||
| No | 99 | 5.5 | 14 | 8.2 | 1.00 | Referent | 1.00 | Referent |
| Yes | 1,713 | 94.5 | 156 | 91.8 | 0.64 | 0.36, 1.15 | 0.59 | 0.31, 1.09 |
| Total | 1,812 | 170 | ||||||
| No. of siblings | ||||||||
| 1 | 534 | 31.2 | 61 | 39.1 | 1.00 | Referent | 1.00 | Referent |
| 2 | 501 | 29.2 | 49 | 31.4 | 0.86 | 0.58, 1.27 | 0.96 | 0.63, 1.45 |
| ≥3 | 678 | 39.6 | 46 | 29.5 | 0.59 | 0.40, 0.89 | 0.83 | 0.54, 1.27 |
| Total | 1,713 | 156 | ||||||
| No. of older siblings | ||||||||
| 0 | 637 | 37.3 | 69 | 44.2 | 1.00 | Referent | 1.00 | Referent |
| ≥1 | 1,072 | 62.7 | 87 | 55.8 | 0.75 | 0.54, 1.04 | 0.88 | 0.61, 1.27 |
| Total | 1,709 | 156 | ||||||
| History of cancer | ||||||||
| No | 1,729 | 95.1 | 168 | 98.2 | 1.00 | Referent | 1.00 | Referent |
| Yes | 89 | 4.9 | 3 | 1.8 | 0.35 | 0.11, 1.11 | 0.52 | 0.16, 1.71 |
| Total | 1,818 | 171 | ||||||
Abbreviations: ABS, address-based sampling; AOR, adjusted odds ratio; CI, confidence interval; OR, odds ratio; RDD, random-digit dialing.
a OR comparing eligible controls recruited by landline and cell-phone RDD with those recruited by ABS.
b OR comparing eligible controls recruited by landline and cell-phone RDD with those recruited by ABS, adjusted for the design variables of age, race, and geographical region; the OR for number of older siblings was further adjusted for number of siblings.
c Region 1: Philadelphia County, Pennsylvania; region 2: Bucks County, Pennsylvania; Chester County, Pennsylvania; Delaware County, Pennsylvania; and Montgomery County, Pennsylvania; region 3: Berks County, Pennsylvania; Lancaster County, Pennsylvania; Lehigh County, Pennsylvania; and Northampton County, Pennsylvania; region 4: Burlington County, New Jersey; Camden County, New Jersey; and Gloucester County, New Jersey.
d Regular: twice a year or more (RDD survey only), once every 6 months or more (ABS survey only), or once a year; irregular: once every 2 years, less than every 2 years, only when sick (RDD survey only), or no primary-care physician.
Table 4.
Characteristics of Enrolled Controls Recruited by Address-based Sampling Compared With Those Recruited by Random-digit Dialing, Philadelphia, Pennsylvania, 2009–2012
| Characteristic | ABS |
RDD |
ORa | 95% CI | AORb | 95% CI | ||
|---|---|---|---|---|---|---|---|---|
| No. | % | No. | % | |||||
| Age group, years | ||||||||
| 18–29 | 157 | 15.4 | 35 | 28.7 | 1.00 | Referent | ||
| 30–39 | 264 | 25.9 | 45 | 36.9 | 0.77 | 0.48, 1.25 | ||
| 40–49 | 308 | 30.2 | 36 | 29.5 | 0.53 | 0.32, 0.87 | ||
| 50–55 | 292 | 28.6 | 6 | 4.9 | 0.09 | 0.04, 0.23 | ||
| Total | 1,021 | 122 | ||||||
| Race | ||||||||
| Nonwhite | 107 | 10.5 | 6 | 4.9 | 1.00 | Referent | ||
| White | 914 | 89.5 | 116 | 95.1 | 2.26 | 0.97, 5.27 | ||
| Total | 1,021 | 122 | ||||||
| Geographical regionc | ||||||||
| Region 1 | 154 | 15.1 | 17 | 13.9 | 1.00 | Referent | ||
| Region 2 | 426 | 41.7 | 75 | 61.5 | 1.59 | 0.91, 2.79 | ||
| Region 3 | 301 | 29.5 | ||||||
| Region 4 | 140 | 13.7 | 30 | 24.6 | 1.94 | 1.03, 3.67 | ||
| Total | 1,021 | 122 | ||||||
| Health status | ||||||||
| Excellent | 259 | 25.4 | 19 | 35.2 | 1.00 | Referent | 1.00 | Referent |
| Very good | 480 | 47.1 | 22 | 40.7 | 0.63 | 0.33, 1.18 | 0.68 | 0.35, 1.32 |
| Good | 230 | 22.6 | 9 | 16.7 | 0.53 | 0.24, 1.20 | 0.80 | 0.34, 1.87 |
| Fair/poor | 51 | 5.0 | 4 | 7.4 | 1.07 | 0.35, 3.27 | 1.83 | 0.54, 6.21 |
| Total | 1,020 | 54 | ||||||
| Primary-care utilizationd | ||||||||
| Regular | 614 | 60.1 | 71 | 58.2 | 1.00 | Referent | 1.00 | Referent |
| Irregular | 361 | 35.4 | 46 | 37.7 | 1.10 | 0.74, 1.63 | 0.95 | 0.63, 1.45 |
| Never | 46 | 4.5 | 5 | 4.1 | 0.94 | 0.36, 2.44 | 0.92 | 0.33, 2.54 |
| Total | 1,021 | 122 | ||||||
| Ever having fathered a child | ||||||||
| No | 353 | 34.6 | 38 | 31.2 | 1.00 | Referent | 1.00 | Referent |
| Yes | 668 | 65.4 | 84 | 68.9 | 1.17 | 0.78, 1.75 | 2.43 | 1.43, 4.12 |
| Total | 1,021 | 122 | ||||||
| Having siblings | ||||||||
| No | 49 | 4.8 | 3 | 5.7 | 1.00 | Referent | 1.00 | Referent |
| Yes | 970 | 95.2 | 50 | 94.3 | 0.84 | 0.25, 2.79 | 0.73 | 0.21, 2.55 |
| Total | 1,019 | 53 | ||||||
| No. of siblings | ||||||||
| 1 | 324 | 33.4 | 26 | 52.0 | 1.00 | Referent | 1.00 | Referent |
| 2 | 280 | 28.9 | 14 | 28.0 | 0.62 | 0.32, 1.22 | 0.63 | 0.31, 1.26 |
| ≥3 | 366 | 37.7 | 10 | 20.0 | 0.34 | 0.16, 0.72 | 0.49 | 0.22, 1.07 |
| Total | 970 | 50 | ||||||
| No. of older siblings | ||||||||
| 0 | 372 | 38.4 | 29 | 58.8 | 1.00 | Referent | 1.00 | Referent |
| ≥1 | 596 | 61.6 | 21 | 41.2 | 0.45 | 0.25, 0.80 | 0.58 | 0.31, 1.10 |
| Total | 968 | 50 | ||||||
| History of cancer | ||||||||
| No | 965 | 94.7 | 53 | 98.2 | 1.00 | Referent | 1.00 | Referent |
| Yes | 54 | 5.3 | 1 | 1.9 | 0.34 | 0.05, 2.48 | 0.49 | 0.06, 3.79 |
| Total | 1,019 | 54 | ||||||
Abbreviations: ABS, address-based sampling; AOR, adjusted odds ratio; CI, confidence interval; OR, odds ratio; RDD, random-digit dialing.
a OR comparing enrolled controls recruited by landline and cell-phone RDD with those recruited by ABS.
b OR comparing enrolled controls recruited by landline and cell-phone RDD with those recruited by ABS, adjusted for the design variables of age, race, and geographical region; the OR for number of older siblings was further adjusted for number of siblings.
c Region 1: Philadelphia County, Pennsylvania; region 2: Bucks County, Pennsylvania; Chester County, Pennsylvania; Delaware County, Pennsylvania; and Montgomery County, Pennsylvania; region 3: Berks County, Pennsylvania; Lancaster County, Pennsylvania; Lehigh County, Pennsylvania; and Northampton County, Pennsylvania; region 4: Burlington County, New Jersey; Camden County, New Jersey; and Gloucester County, New Jersey.
d Regular: twice a year or more (RDD survey only), once every 6 months or more (ABS survey only), or once a year; irregular: once every 2 years, less than every 2 years, only when sick (RDD survey only), or no primary-care physician.
An additional comparison was made between the health status of men aged 18–55 years living in the Northeast, as reported in the National Health Interview Survey, and that of eligible and enrolled controls. Differences between responses from the National Health Interview Survey and those from the ABS group (P = 0.11) or the RDD group (P = 0.32) for eligible controls were not statistically significant, nor were those from the ABS group (P = 0.31) or RDD group (P = 0.41) for enrolled controls.
In order to explore differences in phone usage among persons recruited through ABS and RDD, we included in the study questionnaire questions eliciting information on use of a cell phone or landline or both and percentage of phone usage spent on a cell phone for persons reporting use of both types of phones. Enrolled men with testicular cancer (cases), who were asked the same questions, were used as a comparison group. After adjustment for age, ABS controls were more likely to use cell phones only (AOR = 2.42, 95% CI: 1.35, 4.36) or both type of phones (AOR = 1.87, 95% CI: 1.08, 3.24) than were cases, although there was no difference in the percentage of time spent on a cell phone (Table 5). Controls recruited by landline RDD could not be cell-phone-only users by definition; otherwise, the comparison of this group with cases revealed no difference in landline-only use or both landline and cell-phone use. Among cases and controls recruited via ABS, younger respondents were more likely to use a cell phone (P < 0.001) and to use a cell phone more often than a landline phone (n = 1,074; Spearman's rank correlation coefficient: r = −0.26; P = <0.01) than were older respondents.
Table 5.
Phone-usage Characteristics of Cases Compared With Enrolled Controls, Philadelphia, Pennsylvania, 2009–2012
| Characteristic | Cases |
Controls |
||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| ABS |
Landline RDD |
|||||||||||||
| No. | % | No. | % | ORa | 95% CI | AORb | 95% CI | No. | % | ORc | 95% CI | AORd | 95% CI | |
| Type of phone used | ||||||||||||||
| Landline only | 27 | 4.6 | 30 | 3.0 | 1.00 | Referent | 1.00 | Referent | 5 | 4.5 | 1.00 | Referent | 1.00 | Referent |
| Cell phone only | 152 | 25.8 | 247 | 24.5 | 1.46 | 0.84, 2.55 | 2.42 | 1.35, 4.36 | ||||||
| Both | 410 | 69.6 | 733 | 72.6 | 1.61 | 0.94, 2.74 | 1.87 | 1.08, 3.24 | 106 | 95.5 | 1.40 | 0.53, 3.71 | 1.30 | 0.48, 3.05 |
| Total | 589 | 1,010 | 111 | |||||||||||
| % of time spent on a cell phonee | ||||||||||||||
| 1–24 | 80 | 23.2 | 185 | 25.4 | 1.00 | Referent | 1.00 | Referent | 32 | 30.2 | 1.00 | Referent | 1.00 | Referent |
| 25–49 | 65 | 18.8 | 111 | 15.2 | 0.74 | 0.49, 1.11 | 0.81 | 0.53, 1.22 | 17 | 16.0 | 0.65 | 0.33, 1.28 | 0.63 | 0.31, 1.25 |
| 50–74 | 87 | 25.2 | 189 | 25.9 | 0.94 | 0.65, 1.35 | 1.07 | 0.73, 1.56 | 28 | 26.4 | 0.80 | 0.45, 1.45 | 0.67 | 0.36, 1.23 |
| 75–99 | 113 | 32.8 | 244 | 33.5 | 0.93 | 0.66, 1.32 | 1.32 | 0.91, 1.91 | 29 | 27.4 | 0.64 | 0.36, 1.14 | 0.51 | 0.28, 0.94 |
| Total | 345 | 729 | 106 | |||||||||||
Abbreviations: ABS, address-based sampling; AOR, adjusted odds ratio; CI, confidence interval; OR, odds ratio; RDD, random-digit dialing.
a OR comparing enrolled controls recruited by ABS with cases.
b OR comparing enrolled controls recruited by ABS with cases, adjusted for age.
c OR comparing enrolled controls recruited by RDD with cases.
d OR comparing enrolled controls recruited by RDD with cases, adjusted for age.
e Percentage of total phone usage.
DISCUSSION
Our results demonstrate that ABS can be a useful method of recruiting younger, primarily white males as controls for a case-control study. Initially, we employed landline RDD to recruit controls, resulting in a response rate comparable to rates seen in previous case-control studies (2). However, we perceived the yield for eligible respondents, particularly those youngest in the age distribution, to be too low to allow for effective control recruitment by this method. As a consequence, we piloted the use of RDD incorporating cell-phone numbers, as well as ABS, in an attempt to capture younger populations, who may use only cell phones and whom landline RDD may have missed.
Our findings suggest that groups recruited by means of RDD and ABS were generally comparable with regard to characteristics for which we collected data and for which comparison was appropriate. Differences between RDD and ABS groups in terms of age, race, and geographical region were expected, since respondents were selected on the basis of these characteristics during RDD screening, resulting in artificial differences between groups. Descriptive comparisons of other characteristics between RDD and ABS groups showed few statistically significant differences after adjustment for age, race, and geographical location. Respondents recruited by RDD were more likely to report having fathered a child, and this difference probably reflects landline usage. Young men who use landlines are more likely to have ever fathered a child than are men who use only cell phones (3). Landline use is required for recruitment by RDD but was reported by only 77% of those recruited by ABS. As well, young men aged 18–29 years recruited by RDD were 8-fold more likely to report fair or poor health. However, overall, controls recruited for our Philadelphia-based study were comparable to a representative sample of men aged 18–55 years from the US Northeast in at least 1 respect: health status (15). This result provides some preliminary evidence that our sample may be representative of the stratum of the general population with these age, gender, and geographical characteristics, despite the observed response rates.
ABS and RDD are inherently very different recruitment methods, and consequently it is difficult to generate a suitable measure by which to compare them. We provide some direct comparison using response rates, but these comparisons are made with caution. One concern about this measure is the difference between the denominators for the screening response rate calculations for ABS and RDD. For ABS, the denominator corresponds to all surveys sent out, while for RDD, it consists of all eligible sampling units. It was not possible to determine how many ABS surveys were sent to ineligible sampling units such as addresses with no occupant, although this number is likely to have been small. More importantly, because the text used in the introductory letter that accompanied the ABS screening survey targeted our desired study population (men between the ages of 18 and 55 years), it was not possible for us to fully enumerate persons outside of these limits who would have participated but were subsequently deemed ineligible to participate. The inability to quantitate this group resulted in an artificially low response rate for ABS. Moreover, we did not attempt multiple mailings to individual addresses, which has been shown to increase participant response (9).
Regarding the feasibility of cell-phone RDD, we found that cell-phone RDD was able to reach a higher percentage of unscreened persons of any age than landline RDD, but we were unable to directly determine whether it preferentially reached younger men, as we sought, because of the small sample size of the cell-phone RDD recruitment population. With a larger sample size, it is likely that cell-phone RDD would have captured cell-phone-only users, who tend to be younger, but it is difficult to infer expected numbers based on our data.
Notably, the landline RDD costs for this project per completed participant were roughly double the cost of ABS, while the cell-phone RDD costs were roughly triple. The landline RDD costs were high because this frame does not reach the many young men who do not have landline phones. Cell-phone dialing is expensive because cell-phone numbers must be dialed manually to comply with federal law; cell-phone samples cannot be drawn in a way that effectively targets small geographical areas; and screening is done at the individual level, not the household level (like landline RDD or ABS), so more calls need to be made. This last reason is especially important—approximately half of cell-phone users are female, but these calls to female cell-phone owners did not help us reach our target. Thus, cost-efficiency for cell-phone RDD increases when the target population includes both genders.
The cost differential for ABS versus RDD will depend on factors such as the target population, sample size, incentives, or repeat mailings. Beyond the set-up costs, telephone surveys have fairly constant costs per interview because each interview requires the same amount of interviewer labor. Thus, costs are driven by the number of interviews required to fulfill the study goals. For mail surveys, the cost per completed participant drops substantially with volume because of the economies of scale associated with printing. Thus, costs are driven in part by the expected participant response per bulk mailing and the number of mailings. Furthermore, because ABS reaches potential controls who are cell-phone-only users, this method helps maximize study cost efficiency by allowing retention of cases reporting cell-phone-only use for analyses. Here, the proportion reporting cell-phone-only usage was 25% among controls and 26% among cases.
This study had a number of limitations. Most evident is the low overall response rates, which translates to possible selection bias affecting our study results. While the comparison with National Health Interview Survey data suggests a similarity between study controls and the general population of men in the Northeast, we could only make a comparison with 1 question on general health. Thus, selection bias remains a threat. Additionally, results from the cell-phone RDD screening should be interpreted with caution because of the portability of cell-phone numbers; thus, the cell-phone users selected may not represent the general population of cell-phone users in the Philadelphia area. Finally, our questions probed only a small number of possible respondent characteristics. In particular, we did not ask questions about tobacco use, alcohol consumption, lack of health insurance, or other characteristics that have been shown to have a higher prevalence in cell-phone-only users and which make cell-phone-only users an important subpopulation to include in research (16, 17).
Importantly, the generalizability of the study findings is limited because some respondents were screened out or excluded via quotas based on age, race, and ethnicity early in the RDD screening and were not administered the full screening survey. This selection was intentional, since there is a greater incidence of testicular cancer among younger, white men, and we sought to recruit a control group comparable to our case population. Our sample therefore consisted primarily of white males as compared with persons of other gender, racial, and ethnic groups. Consequently, one cannot necessarily extrapolate these findings to the broader population. We were also unable to directly examine whether cell-phone RDD results in a sample population that is representative of the general population in terms of race and other commonly reported variables. However, other authors have examined this issue elsewhere (18–20). In addition, not all questions were asked during all rounds of screening, which further limited the sample size for responses to questions on health status, likelihood of having siblings, number of siblings, number of older siblings, and history of cancer other than testicular cancer.
In summary, we find ABS to be an adequate method of recruiting younger, primarily white males, which results in a sample population similar to that for RDD. With the increasing trend in cell-phone usage, the effectiveness of landline RDD for control recruitment would appear to be decreasing. Consequently, researchers may consider the use of ABS screening in future case-control studies, particularly those that require recruitment of younger populations. However, special attention is necessary because of low overall response rates using ABS methods.
ACKNOWLEDGMENTS
Author affiliations: Center for Clinical Epidemiology and Biostatistics, Perelman School of Medicine at the University of Pennsylvania, Philadelphia, Pennsylvania (Bartholt Clagett, Stephanie L. Ciosek, Nandita Mitra, Peter A. Kanetsky); Department of Biostatistics and Epidemiology, Perelman School of Medicine at the University of Pennsylvania, Philadelphia, Pennsylvania (Nandita Mitra, Peter A. Kanetsky); Department of Medicine, Division of Translational Medicine and Human Genetics, Perelman School of Medicine at the University of Pennsylvania, Philadelphia, Pennsylvania (Katherine L. Nathanson, Monique McDermoth); Department of Medicine, Division of Hematology/Oncology, Perelman School of Medicine at the University of Pennsylvania, Philadelphia, Pennsylvania (David J. Vaughn); and Abt SRBI, Silver Spring, Maryland (Andrew Weiss, Rachel Martonik).
This work was supported by a grant to K.L.N. from the National Institutes of Health (grant R01CA114478).
We thank Drs. Greta Bunin, Samuel Lesko, and Stephen Schwartz for their contributions as members of the study's advisory committee.
The study sponsor was not involved in the design of the study, the data analyses, or manuscript preparation.
Andrew Weiss and Rachel Martonik are employees of Abt SRBI (Silver Spring, Maryland). Abt SRBI completed all landline and cell-phone RDD recruitment and the pilot phase of ABS recruitment for this study.
REFERENCES
- 1.Olson SH, Kelsey JL, Pearson TA, et al. Evaluation of random digit dialing as a method of control selection in case-control studies. Am J Epidemiol. 1992;135(2):210–222. doi: 10.1093/oxfordjournals.aje.a116273. [DOI] [PubMed] [Google Scholar]
- 2.Bunin GR, Spector LG, Olshan AF, et al. Secular trends in response rates for controls selected by random digit dialing in childhood cancer studies: a report from the Children's Oncology Group. Am J Epidemiol. 2007;166(1):109–116. doi: 10.1093/aje/kwm050. [DOI] [PubMed] [Google Scholar]
- 3.Voigt LF, Schwartz SM, Doody DR, et al. Feasibility of including cellular telephone numbers in random digit dialing for epidemiologic case-control studies. Am J Epidemiol. 2011;173(1):118–126. doi: 10.1093/aje/kwq322. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Blumberg SJ, Luke JV. Wireless Substitution: Early Release of Estimates From the National Health Interview Survey, January–June 2012. Hyattsville, MD: National Center for Health Statistics; 2012. http://www.cdc.gov/nchs/data/nhis/earlyrelease/wireless201212.PDF. ). (Accessed April 12, 2013) [Google Scholar]
- 5.Iannacchione VG. The changing role of address-based sampling in survey research. Public Opin Q. 2011;75(3):556–575. [Google Scholar]
- 6.Boyle J, Fleeman A, Kennedy C, et al. Sampling cell phone only households: a comparison of demographic and behavioral characteristics from ABS and cell phone samples. Surv Pract. 2012;5(1):1–7. [Google Scholar]
- 7.Iannacchione VG, Staab JM, Redden DT. Evaluating the use of residential mailing addresses in a metropolitan household survey. Public Opin Q. 2003;67(2):202–210. [Google Scholar]
- 8.Link MW, Battaglia MP, Frankel MR, et al. Address-based versus random-digit-dial surveys: comparison of key health and risk indicators. Am J Epidemiol. 2006;164(10):1019–1025. doi: 10.1093/aje/kwj310. [DOI] [PubMed] [Google Scholar]
- 9.Link MW, Battaglia MP, Frankel MR, et al. A comparison of address-based sampling (ABS) versus random-digit dialing (RDD) for general population surveys. Public Opin Q. 2008;72(1):6–27. [Google Scholar]
- 10.Johnson P, Williams D. Comparing ABS vs. landline RDD sampling frames on the phone mode. Surv Pract. 2010 June. http://surveypractice.wordpress.com/2010/06/22/abs-vs-landline-rdd/ ). (Accessed September 25, 2012) [Google Scholar]
- 11.Waksberg J. Sampling methods for random digit dialing. J Am Stat Assoc. 1978;73(361):40–46. [Google Scholar]
- 12.Bevier M, Weires M, Thomsen H, et al. Influence of family size and birth order on risk of cancer: a population-based study. BMC Cancer. 2011;11:163. doi: 10.1186/1471-2407-11-163. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Olson SH, Voigt LF, Begg CB, et al. Reporting participation in case-control studies. Epidemiology. 2002;13(2):123–126. doi: 10.1097/00001648-200203000-00004. [DOI] [PubMed] [Google Scholar]
- 14.Slattery ML, Edwards SL, Caan BJ, et al. Response rates among control subjects in case-control studies. Ann Epidemiol. 1995;5(3):245–249. doi: 10.1016/1047-2797(94)00113-8. [DOI] [PubMed] [Google Scholar]
- 15.National Center for Health Statistics. National Health Interview Survey, 2009 (ICPSR 28721) Ann Arbor, MI: Inter-University Consortium for Political and Social Research; 2010. (Version 2) (http://www.icpsr.umich.edu/icpsrweb/ICPSR/studies/28721. ). (Accessed July 20, 2012) [Google Scholar]
- 16.Currivan D, Roe D, Stockdale J. The Impact of Landline and Cell Phone Usage Patterns Among Young Adults on RDD Survey Outcomes; Research Triangle Park, NC: RTI International; 2008. (http://www.rti.org/pubs/currivan_aapor08_paper.pdf. ). (Accessed September 25, 2012) [Google Scholar]
- 17.Delnevo CD, Gundersen DA, Hagman BT. Declining estimated prevalence of alcohol drinking and smoking among young adults nationally: artifacts of sample undercoverage? Am J Epidemiol. 2008;167(1):15–19. doi: 10.1093/aje/kwm313. [DOI] [PubMed] [Google Scholar]
- 18.Kalton G, Piesse A. Survey research methods in evaluation and case-control studies. Stat Med. 2007;26(8):1675–1687. doi: 10.1002/sim.2796. [DOI] [PubMed] [Google Scholar]
- 19.Puumala SE, Spector LG, Robison LL, et al. Comparability and representativeness of control groups in a case-control study of infant leukemia: a report from the Children's Oncology Group. Am J Epidemiol. 2009;170(3):379–387. doi: 10.1093/aje/kwp127. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Wang PP, Dicks E, Gong X, et al. Validity of random-digit-dialing in recruiting controls in a case-control study. Am J Health Behav. 2009;33(5):513–520. doi: 10.5993/ajhb.33.5.4. [DOI] [PubMed] [Google Scholar]