Real-Time Data Collection Using Text Messaging in a Primary Care Clinic

Introduction

Text messaging is widely used across nearly all age and income categories. As of March 2015, 92% of all US adults own a cell phone, and at least 81% of all cell phone owners use their device for texting. ^1,2 Thus far, text messaging has been successfully used in healthcare to promote behavioral change and improve disease management (e.g., by encouraging smoking cessation, improving medication compliance, increasing vaccination rates, disseminating accurate health information, and facilitating weight loss). ^{3 –14}

In addition to supporting health-related behavioral change, text messaging is a promising means of eliciting patient perspectives. Patients could potentially use text messaging to “speak” to researchers about their healthcare knowledge, opinions, and beliefs. ^15,16 Such a patient-centered data collection method shows promise for several reasons.

Text message surveys have been found to be cost-effective and easy to administer in low-income, community-based settings. ¹⁵ In addition, they offer anonymity and the convenience of asynchronous communication (so patients can respond to questions when it is convenient for them). Texting is a familiar and ubiquitous mode of communication, and may be particularly advantageous for reaching vulnerable populations such as minorities and adolescents. ^17,18 In fact, a greater proportion of blacks and Hispanics report that they use text messaging compared with whites (85% and 87%, compared with 79%, respectively). ¹⁷ Moreover, they offer the advantage of collecting data in real-time, enabling assessment of evolving patient needs and preferences, and facilitating the translation of collected data into timely healthcare and policy change. When compared with Web-based surveys that can be completed through a smartphone (i.e., through mobile apps), text message surveys offer the advantage of not requiring Internet access and thus including people who do not have a smartphone or are conscious of their mobile data usage.

The use of text messaging as a two-way channel of communication between researcher and participant has thus far ranged from simple, numerical data collection (i.e., asking patients to report their blood glucose levels) ^{19 –21} to more complex clinical interaction (i.e., patients conversing back and forth with pharmacists regarding their medication questions). ²² A few studies ^{23 –27} have specifically used text message surveys, but largely to collect data before enrollment into clinical trials ^{23 –26} or via unsolicited “text blasts” to large groups, which has yielded very low (3.6%) response rates. ²⁷ However, text messaging has yet to be used to collect real-time patient opinions and perspectives to improve clinical care and inform policy issues. The aim of this study was to assess the feasibility and acceptability of administering a large-scale text message survey and to describe lessons learned that may inform future text message surveys among clinical populations.

Materials and Methods

Results

Our convenience sample consisted of 461 individuals who attempted the survey. Of those individuals, 80% (370/461) completed it in full. Respondents were predominantly female, non-Hispanic black or non-Hispanic white, and had at least some college education (Table 1).

The clinic from which we sampled is located in a community where 33% of the population is in poverty. ³⁰ The demographics of the community are comparable with the demographics of our study, with greater representation from females and black community members (Community vs. Clinic Sample): 50% versus 75% female, 29% versus 42% black, 59% versus 41% white, 4% versus 4% Hispanic, 33% versus 39% Bachelor degree or higher.

Of those who completed the survey, 84% (312/370) reported willingness to participate in a hypothetical future texting survey for a gift card (with no monetary amount specified). People who attempted but did not complete the survey (91/461) answered a median of 9 (a mean of 9.3) questions before they stopped responding.

Smartphone users were less likely to leave the survey unfinished compared with flip phone (or “other” type of phone) users (15% vs. 30%, respectively; p = 0.004). In comparing these two populations, there was no significant difference between their age or gender. However, smartphone users were more likely to be non-Hispanic white (p = 0.032) and more educated (p = 0 02). Among those who completed the survey (n = 370), the median time from initiation to completion was 14 min (range: 3 min to 22 h), with 71% completing the survey within 30 min. Among those who attempted but did not complete the survey (n = 91), the median time spent interacting with the survey was 8 min (range: 0 min to 18 h 33 min).

In each of the 91 instances where a user started but did not complete the survey, investigators analyzed the text message exchange between the two parties. The content and duration of time between each message was noted. In 73% (66/91) of these exchanges, no clear reason for leaving the survey could be extrapolated from the text history. In 4% (4/91) of the exchanges, the user submitted a response that contained the word “stop,” as in “she should stop smoking.” By doing so, the user inadvertently opted out of all further survey questions. In 23% (21/91) of the exchanges, frustration with error messages may have led to the abandonment of the survey. One demonstrative excerpt from a real text exchange is presented in Figure 1. Additional examples can be found in Appendices 2–4.

OPEN IN VIEWER

Fig. 1.

Example of text message exchange that lead to respondent attrition.

Discussion

Consistent with previous community-based samples, ¹⁵ our findings demonstrate that both quantitative and qualitative data can be effectively collected via text message from a diverse clinic-based sample. Over 450 respondents were recruited over a 32-day period for only a nominal incentive, which represents a significantly lower financial burden than other forms of survey data collection. ^31,32 In addition, the large proportion of respondents that would complete another survey demonstrates the appeal of this method of data collection among our sample.

Despite their ease of implementation, text message surveys also represent unique challenges compared with other forms of survey data collection. Several important lessons emerged from our experiences that may inform the future use of this simple technology for clinical research:

(1) Recruit participants in a setting where there is an opportunity to complete the survey immediately. Although respondents could answer the survey at any time, the vast majority of those that completed the survey did so within the first few minutes of being in the waiting room. The waiting room of a medical clinic was ideal in this regard, as patients often had unoccupied “down time.”

Conclusions

Text-message surveys are a feasible and acceptable method to collect real-time data among low-income, clinic-based populations. Furthermore, our findings provide valuable insights into this novel survey modality and demonstrate its broad applicability in collecting patient-centered data to improve the efficacy and impact of clinical interventions and policies.