We analyzed secondary, cross-sectional data from the 2011-2015 NHIS. Since 1957, the NHIS has collected information on demographics, socioeconomics, and a wide range of health topics among the civilian noninstitutionalized US population [28]. During 2011-2015, the NHIS used a multistage sampling design to monitor national trends in health, illness, and disability while tracking progress toward national goals by using a computer-assisted personal interviewing system during face-to-face interviews [28]. The sampling design oversampled Hispanic, Asian, and non-Hispanic Black persons to increase the precision of estimates among racial and ethnic minorities [28]. Information about health information technology use and cancer screening behaviors have been measured annually since 2011 [29-31]. Further details of the NHIS sampling design and data collection methods have been reported previously [28].

We limited our sample to individuals who received primary health care in the past 12 months and were within the recommended age groups for each cancer screening of interest. For breast cancer screening, for instance, our sample was limited to women aged 40 years and above without any prior history of breast cancer based on American College of Obstetricians and Gynecologists (ACOG) screening recommendations (n=35,912) [32]. For cervical cancer screenings, our sample was limited to women aged 21-65 years without any prior history of cervical cancer based on the American Cancer Society (ACS), ACOG, and the United States Preventive Services Task Force (USPSTF) screening recommendations (n=48,512) [33-35]. For colon cancer screenings, our sample was limited to adults aged 50 years and older without any prior history of colon cancer based on the ACS screening recommendations (n=45,884) [33]. Individuals who reported that they do not use the internet were excluded.

Bivariate analyses were used to describe the association between demographic, socioeconomic, health-related characteristics, use of email PPC, and receipt of each cancer screening in the past 12 months (chi-square test; =.05). We calculated age-adjusted prevalence estimates of receiving breast, cervical, and colon cancer screenings among adults who reported using email PPC by using estimated marginal (least-squares) means. Crude and multivariable logistic regression procedures were used to test for associations between email PPC (independent variable) and whether they received breast, cervical, and colon cancer screenings (dependent variable) before and after controlling for covariates. Purposeful selection methods were used for building fitted multivariable logistic regression models [36]. Our multivariable models were adjusted for age (reference: youngest age group, 40-49 years for breast cancer screenings, 21-30 years for cervical cancer screenings, and 50-59 years for colon cancer screenings); race or ethnicity (reference: non-Hispanic White); marital status (reference: never married); education (reference: bachelor’s degree or higher); health insurance (reference: covered); and perceived health status (reference: fair or poor). For colon cancer screenings only, multivariable models were adjusted for sex (reference: men).

Sensitivity analyses were conducted to align with different age cutoffs for screening recommendations from other agencies based on previous research [37]. For breast cancer and colon cancer screenings, we limited the sample to women aged 50-75 years based on USPSTF recommendations [37,38]. No sensitivity analyses were conducted for cervical cancer screenings.

Data were analyzed using SAS software (version 9.4) survey procedures to account for primary sampling units, clustering, and the sophisticated weighting in the sampling design. The annual sample adult weight was divided by five to account for combining 5 years of data based on NHIS analytic recommendations [39].

This study was deemed exempt from human subjects review by the Committee for the Protection of Human Subjects at the University of Texas Health Science Center at Houston.

The mean ages of women who received breast and cervical cancer screenings in the past 12 months were 56.90 years (95% CI 56.71-57.08) and 41.2 (95% CI 41.00-41.40) years, respectively. The mean age of adults who received a colon cancer screening the past 12 months was 62.1 (95% CI 61.94-62.35) years. Women who reported having a breast cancer screening in the past 12 months were more likely to be non-Hispanic White; be married or living with a partner; have a bachelor’s degree or higher level of education; have health insurance coverage; and perceive their health as excellent, very good, or good (all P<.05). Similar results were observed for cervical cancer screenings. Additionally, US-born women (27,006/31,977, 85.5% weighted) were more likely to receive a cervical cancer screening than foreign-born women (4971/31,977, 14.5% weighted; P<.001). Furthermore, over half (5587/11,713, 50.9% weighted) of the adults who received a colon cancer screening were male (P<.001). Results were similar to those for breast and cervical cancer screenings for race or ethnicity, marital status, education, and health insurance coverage (all P<.001). Further details of the bivariate analyses (unweighted frequencies and weighted percentages) are reported in Table 1.

Age-adjusted prevalence estimates of receiving a breast, cervical, or colon cancer screening based on email PPC use are reported in Table 2. Among adults who received primary health care in the last 12 months, the age-adjusted prevalence of receiving a colon cancer screening was the lowest (34.4%) among those who used email PPC compared to women who underwent breast (70.5%) and cervical (70.6%) cancer screenings.

Crude and adjusted logistic regression results are reported in Table 3. In adjusted models, women who used email to communicate with their health care providers had 1.32 times greater odds (95% CI 1.20-1.44) of receiving a breast cancer screening and 1.11 times greater odds (95% CI 1.02-1.20) of receiving a cervical cancer screening than women who did not use email PPC. Adults who used email to communicate with their health care providers had 1.55 times greater odds (95% CI 1.42-1.69) of receiving a colon cancer screening than those who did not use email PPC. Specific estimates for covariates included in our logistic regression models are provided in Table S1 of Multimedia Appendix 1

Crude and adjusted logistic regression results from our sensitivity analysis are reported in Table S2 of Multimedia Appendix 1. The results were similar to our analytical sample. All 95% CIs overlapped with our initial findings.

We aimed to determine the association between email PPC and whether adults received breast, cervical, and colon cancer screenings. Overall, we found that adults who used email to communicate with their health care providers between visits had greater odds of receiving each of the three types of screenings. These findings go beyond our previous research that used other nationally representative data sources (Medical Expenditure Panel Survey and HINTS), which demonstrated that quality face-to-face PPC increased the likelihood of adults receiving cancer screenings [15,16]. Nevertheless, there was no difference in breast, cervical, or colon cancer screening uptake among adults who did and those who did not use email PPC.

Using the NHIS, we were able to further explore the role of email PPC as a marker of the likelihood of adults receiving cancer screening using a nationally representative sample larger than that used in previous studies. For breast cancer screening, we found that women who used email PPC had 32% increased odds of receiving a mammogram compared to women who did not use email PPC. Other studies exploring whether online PPC and general health information technology use were associated with breast cancer screening found that electronic medical record and patient portal use increased women’s likelihood of receiving mammograms [23,40-42]. Moreover, we found that email PPC increased women’s odds of receiving a Pap test by 11%. This finding result differs from our study using HINTS data, which did not find any association between email PPC and cervical cancer screenings [15]. To our knowledge, only one other study has demonstrated that general electronic medical record use can increase cervical cancer screenings [41]. Finally, we found that the use of email PPC increased the likelihood of adults receiving a colon cancer screening by 55%. Our previous study using HINTS data indicated that adults had 39% higher odds of receiving a colon cancer screening; however, the results did not reach statistical significance (95% CI 0.99-1.95). Similar to studies evaluating breast cancer screening, previous research has demonstrated that adults who used patient portals to schedule appointments, request referrals or prescription refills, view decision aids, share medical records, or communicate with health care professionals by email were more likely to receive colon cancer screenings than those who did not use patient portals [22,42]. For all cancer screenings, the lack of research providing direct comparisons to our results may be due to limitations of examining email PPC only, which excludes other online communicative functions such as text messaging, mobile apps, and social media [43].

For all cancer screenings, several factors may have contributed to obtaining results different from our previous study using HINTS data [15]. The greatest factor may be the way email PPC was measured. The NHIS measured whether adults used computers or the internet to communicate with their health care provider by email [28]. A similar measure was used in the HINTS 4 survey during Cycles 1 and 3 [44]. Adults who responded “yes” on either survey may have regarded automatic emails for appointment reminders or diagnostic test results as email PPC versus directly emailing their health care provider about specific health concerns. During HINTS Cycle 3 and 4, this question was revised to directly assess whether adults exchanged health information with their health care provider via email [44]. The reasoning for and quality of communication remained unmeasured by both surveys. Future iterations of these data sources should be revised to fully capture communication behaviors to further explore the implications of email PPC on the uptake of preventive services.

A strength of this study was the use of survey data from multiple years of the NHIS, a nationally representative survey that has consistently measured health behaviors, preventive health services, and a wide array of other health-related characteristics to meet national health objectives for over 60 years [28]. The depth and breadth of demographic, socioeconomic, and health-related characteristics measured by the NHIS on an annual basis allowed us to explore and control for multiple covariates in our logistic regression models. However, the NHIS does not collect characteristics on patient engagement in health care outside from assessing adults’ use of health information technology to look up health information, refill prescriptions, schedule appointments, use online chat groups, and communicate with health care providers via email that may result in unresolved confounding. Previous research has demonstrated that adults who use electronic methods of communication with their health care providers and adults who follow recommended cancer screening guidelines are more engaged in their health care than those who do not [5]. In our study, email PPC and cancer screening behaviors were both measured in the past 12 months based on how the questions were asked by the NHIS. Our results may suffer from temporality biases, as we were unable to determine whether email PPC occurred before or after receiving any cancer screening. By limiting our cancer screening outcome to the past 12 months instead of based on adherence (eg, past 2 years for mammogram, past 3 years for Pap testing, and past 10 years for colonoscopy), we minimized this potential bias. A limitation to our measurement of email PPC was that we were unable to determine the direction of the email (ie, patient to provider vs provider to patient) and whether the content of the communication was screening related or pertaining to any medical information. The clinical significance of our results may be limited due to reporting odds ratios and marginal means. The use of other marginal effects may have improved our study’s practical relevance [45]. Our cancer screening measures were self-reported. Some studies have cautioned that results from self-reported nationally representative studies may overestimate cancer screening uptake [46], whereas others have found that self-reported responses are consistent with findings from hospital records [47,48]. It is also important to note that the NHIS is a cross-sectional survey, so our results only represent associations instead of causal relations.

This study begins to fill the gap in our understanding of how email PPC and direct electronic messaging between appointments may be a marker of the increased likelihood of adults receiving preventive health services, in particular, cancer screening uptake. More research is needed to determine the need for and effectiveness of targeted strategies for promoting appropriately timed cancer screenings by using web-based PPC tools such as email and direct messaging. Furthermore, there is a need for more research to examine reasons for and quality of email PPC for making preventive health care decisions.