Factors Associated With Opting Out of Automated Text and Telephone Messages Among Adult Members of an Integrated Health Care System

This cohort study assesses whether a high volume of automated text or interactive voice response telephone messages is associated with requests to opt out of future messages.


Introduction
Reminding patients of upcoming clinical appointments, laboratory monitoring, and other scheduled services may be associated with improved efficiency and effectiveness of health care. Systematic reviews have shown that reminders are associated with reduced missed primary and specialty care appointments, increased delivery of preventive services, improved prescription refill rates, and enhanced laboratory monitoring. [1][2][3] Over time, the technology for delivering reminders has largely shifted from live telephone calls and postal reminders to automated text messages and interactive voice response (IVR) telephone reminders to reduce staff burden and costs. These communication channels have also facilitated campaigns to promote healthy behavior or disseminate information to patients within health care systems. 4 The COVID-19 pandemic has accelerated the reliance of health care systems on virtual communication channels to inform patients about care options and provide guidance about preventive practices. This reliance will likely increase as systems address the backlog of deferred care associated with the pandemic and continue to emphasize virtual care. 5,6 Health care systems usually implement messaging interventions for a specific clinical purpose but rarely assess the aggregate volume of messages delivered to an individual patient over time.
Little research has quantified the cumulative volume of text and IVR messages that patients receive from their health care system or assessed message fatigue, defined as "an aversive motivational state of being exhausted and bored by overexposure to similar redundant messages over a prolonged period of time." 7(p10) In the Kaiser Permanente Colorado (KPCO) health system, randomized clinical trials have demonstrated that text and IVR messages can reduce missed primary care appointments, increase medication refills and laboratory test completion, improve child and adult immunization rates, transmit home blood pressure readings, and facilitate obesity counseling. [8][9][10][11][12][13][14][15] Thus, these communication channels are widely used within the system. In this study, we assessed the total number of text and IVR messages delivered by KPCO to its adult members over a 1-year period and the association between the volume of messages and message fatigue, defined as requests to stop delivery of text or IVR messages.

Methods
We conducted a retrospective cohort study to assess text and IVR message fatigue in KPCO members. The study cohort consisted of all adult KPCO members who received at least 1 text or IVR message between October 1, 2018, and September 30, 2019. These individuals were included whether or not they had in-person visits during the year because all members could receive text or IVR messages. Individuals with KPCO insurance for part of the year were included for their period of enrollment. We excluded individuals who had requested not to participate in medical record-based research or who had special restrictions on access to health records. The Kaiser Permanente Each text message allowed recipients to opt out, defined as a request for discontinuation of all future text messages, by reply text. Each IVR message provided recipients the option to opt out of further messages for that specific clinical purpose.
Although most text and IVR messages were delivered by a regional team, Kaiser Permanente's national program also delivered some text and telephone messages to KPCO members. In addition, KPCO members could communicate with their health care team and receive reminders, updates, clinical information, and other messages through the Kaiser Permanente patient portal. 18 By 2019, 78.6% of KPCO members were enrolled in the patient portal, and estimated email volume exceeded 6.5 million messages per year.
We excluded email communications from this analysis because members voluntarily established and used their account on the patient portal and because email communications might be less likely to be associated with message fatigue than text or IVR messages delivered without formal consent. We also excluded text or IVR communications from the Kaiser Permanente national program because records of these contacts were not available. Telephone reminder calls from KPCO staff to patients were excluded because they were not systematically captured in the KPCO electronic health record. Thus, text and IVR reminders issued by the KPCO regional team underestimated total member outreach through all communication channels.

Study Measures
We defined message volume for each cohort member and each communication channel as the total number of messages that KPCO attempted to deliver during the study year. Text and IVR message volumes were assessed separately. We counted a message even if it was not successfully delivered.
Each message was categorized as an appointment reminder, a medication refill reminder, a screening reminder, a vaccination reminder, a diagnostic test reminder, or another reason for communication.
Messages sent for research studies were excluded because they did not contribute to clinical care. Some individuals in the cohort were also included in 3 large-scale text message campaigns. In March 2019, adults 65 years and older were informed about the accessibility of primary care in Kaiser Permanente. In June and August 2019, members received an anticipatory message encouraging them to seek care from a KPCO facility rather than a non-KPCO site during the upcoming holidays.
These text messages were categorized as text campaigns for this analysis.
We collected information about text and IVR messages from the software systems that delivered the messages (Twilio text message software and Voxeo IVR software). Message variables included a project identifier, the type and date of message (text or IVR), the response (delivered, voicemail, no answer, or busy), and the date and time of the message. The system also recorded whether the member had responded to a text message with a request to stop receiving all future text messages and whether a member had responded to an IVR call with a request to stop receiving future IVR messages for that specific indication. We restricted both message-level and member-level an established set of ICD-10 codes. 20 Rates of use for primary and specialty outpatient care visits, emergency department visits, and hospitalizations were calculated for visits to KPCO facilities and out-of-system services paid for by KPCO.

Statistical Analysis
We conducted separate analyses for text and IVR messages. Individuals who received both text and IVR messages were included in both analyses. We conducted 2-sided bivariate comparisons using t tests for continuous variables and χ 2 tests for categorical variables. For members of the cohort who did not opt out of text and IVR messages, we calculated the rate of messages per year as the number of messages attempted, divided by the proportion in the year during which they were enrolled in the health plan. For individuals who opted out, we calculated the rate of messages per year as the number of messages delivered, divided by the proportion in the year that elapsed before the date of their opt-out request. We used multivariable logistic regression models to examine the association between each outcome (text or IVR) and sociodemographic, insurance, social, and clinical factors.
We did not include service use variables in these analyses because use could be directly associated with the reminder text or call. The discrimination of logistic regression models was evaluated using the C statistic. All analyses were performed using SAS, version 9.4 (SAS Institute Inc). A 2-sided P < .05 was deemed statistically significant.  Figure 1 shows the proportions of cohort members who received no messages, text messages only, IVR messages only, and both text and IVR messages. Overall, 84.1% of cohort members received 1 or more text messages, and 67.8% received 1 or more IVR calls. eTable 2 in the Supplement shows a comparison of characteristics of individuals in the study cohort who received text messages only, IVR calls only, and both types of automated message. Among individuals who received text messages,  Among participants who opted out of text messages, 19.3% had received 20 or more messages, compared with 6.4% of those who did not opt out of text messages. Among individuals who opted out of IVR messages, 33.5% had received 20 or more messages compared with 3.8% of those who did not opt out of IVR messages.

Discussion
In this cohort study, we assessed the proportion of adult members of an integrated health care system who opted out of receiving future automated text messages or IVR telephone calls over a 1-year period. The incidence of opting out over the 1-year period was low for in the overall cohort but increased in association with message volume, particularly among individuals receiving IVR messages. Individuals who opted out of receiving either type of message were substantially more likely to opt out of the other type during the study year. Older individuals were more likely to opt out of either communication channel, whereas other sociodemographic, clinical, or social variables such as sex and race/ethnicity were not consistently associated with opting out. We also observed a transient increase in opt-out requests after text message campaigns that were broadly informational rather than focused on individual health care needs. Message fatigue has long been a concern for marketing and advertising campaigns outside health care. 7 Existing research 7,26,27 has primarily assessed the attitudes expressed by recipients in response to health messages rather than requests to discontinue those messages. These studies have shown that message fatigue may be an unintended consequence of public health messages about safe sex, obesity prevention, and smoking cessation. [26][27][28][29] Message fatigue may also jeopardize the effectiveness of public health messages during the COVID-19 pandemic. 30 The findings of this study should be replicated in other health care systems. If additional research is confirmatory, future qualitative and quantitative studies should explore the complex relationships between patient attitudes about automated messages, their preferences for communication channels and message frequency, and their decisions to continue or opt out of future messages. Our findings suggest that the association of patient age with these attitudes is particularly important because opt-out rates for both text and IVR messages increased as age increased despite accumulating medical and social risk factors and greater use of care.
Despite the low rates of message fatigue that we observed, health care systems should consider approaches to reduce this problem and optimize the effectiveness of their automated communication strategies. [2][3][4] Incorporating patient preferences about the channel, frequency, and topics of communication from the system could ensure that individuals receive the content they desire through the channels they prefer at a frequency they would accept. Awareness of personal preferences could also inform a coordinated governance and prioritization process among multiple entities within the delivery system that generate IVR-, text-, or email-based message campaigns. 31 Messages that are currently delivered separately could be bundled into a single communication that identifies all upcoming health care needs. 29 Predictive models could direct reminder messages about upcoming appointments to individuals at greatest risk of missing those appointments. Research in KPCO and other settings has shown that patients who will likely miss appointments can be predicted with high accuracy using electronic health record variables. 12,13,32 "Blast" messages about topics that are important to the delivery system but have limited personal relevance may have unintended consequences that should be considered before deployment.

Limitations
This study has limitations. The generalizability of our findings is limited by the inclusion of adult members from a single integrated health care system and by restriction of the analysis to a single year. We assessed only text and IVR messages, although email messages, in-person telephone calls, and text or IVR messages from national Kaiser Permanente sources also contributed to total message volume and may have exacerbated message fatigue.
Additional limitations of the study include our inability to assess the number of individuals who experienced message fatigue but did not opt out; a survey would be necessary to assess these attitudes. 26,27 Because the decision to opt out of text messages was associated with cessation of all further texts, whereas the decision to opt out of IVR messages was associated only with cessation of calls for that specific purpose, opt-out rates and patient characteristics may not be comparable between the 2 communication channels. We could not determine why individuals chose to opt out in response to a specific message. Information on member preferences for communication channels was not consistently available.

Conclusions
In this cohort study, a small proportion of adult KPCO members demonstrated message fatigue through requests to discontinue text and IVR messages from the system. The prior volume of

JAMA Network Open | Health Informatics
Factors Associated With Opting Out of Automated Messages in a Health Care System messages received was associated with message fatigue. Recognition of this unintended consequence of automated text and IVR communication may promote message delivery strategies that improve patient-centeredness while preserving the effectiveness of these important tools.