Review of Computer-generated Outpatient Health Behavior Interventions
Clinical Encounters “in Absentia”
- Correspondence and reprints: Debra Revere, MA, MLIS, University of Washington, IAIMS Program, Box 357155, Seattle WA 98195-7155; email: <drevere{at}u.washington.edu>
- Received 16 March 2000
- Accepted 21 August 2000
Abstract
Objective To evaluate evidence of the effectiveness of computer-generated health behavior interventions—clinical encounters “in absentia”—as extensions of face-to-face patient care in an ambulatory setting.
Data Sources Systematic electronic database and manual searches of multiple sources (1996–1999) plus search for gray literature were conducted to identify clinical trials using computer-generated health behavior interventions to motivate individuals to adopt treatment regimens, focusing on patient-interactive interventions and use of health behavior models.
Study Selection Eligibility criteria included randomized controlled studies with some evidence of instrument reliability and validity; use of at least one patient-interactive targeted or tailored feedback, reminder, or educational intervention intended to influence or improve a stated health behavior; and an association between one intervention variable and a health behavior.
Data Extraction Studies were described by delivery device (print, automated telephone, computer, and mobile communication) and intervention type (personalized, targeted, and tailored). We employed qualitative methods to analyze the retrieval set and explore the issue of patientinteractive computer-generated behavioral intervention systems.
Data Synthesis Studies varied widely in methodology, quality, subject number, and characteristics, measurement of effects and health behavior focus. Of 37 eligible trials, 34 (91.9 percent) reported either statistically significant or improved outcomes. Fourteen studies used targeted interventions; 23 used tailored. Of the 14 targeted intervention studies, 13 (92.9 percent) reported improved outcomes. Of the 23 tailored intervention studies, 21 (91.3 percent) reported improved outcomes.
Conclusions The literature indicates that computer-generated health behavior interventions are effective. While there is evidence that tailored interventions can more positively affect health behavior change than can targeted, personalized or generic interventions, there is little research comparing different tailoring protocols with one another. Only those studies using print and telephone devices reported a theoretic basis for their methodology. Future studies need to identify which models are best suited to which health behavior, whether certain delivery devices are more appropriate for different health behaviors, and how ambulatory care can benefit from patients' use of portable devices.
The number of controlled trials, reviews, and reports in the literature and popular media suggests that interest in using technology to augment patient– physician interactions has increased in the last decade. Recently, a JAMIA article recommended that telemedical services and information systems address behavior change, individual risk factors, and patient education, and further predicted that “the trend is toward delivery of care in an ambulatory setting or by interaction with a patient directly at home, and telemedicine services and information systems provide the necessary communication links.”1
The purpose of this study is to report the current state of the peer-reviewed evidence for patient-interactive computer-generated health behavior interventions—clinical encounters “in absentia”—as extensions of face-to-face patient care. We were interested in two specific areas: the health behavior models used in these interventions and the devices used for patient education, counseling, and reminder systems aimed at improving patient health behaviors.
Background
Other Reviews
Other reviews have focused on a specific delivery method (e.g., telephone-delivered interventions) or a particular health behavior focus (e.g., smoking cessation). To our knowledge, this is the first review to summarize findings across all interventions that involve devices that communicate or interact directly with the patient, regardless of technology, health behaviors, or medical conditions. Intervention types are also defined more narrowly and more consistently than in previous literature reviews (as discussed under Methods). While other reviews describe the growing role of telecommunication in health care, this review specifically examines the state of computer-generated or computer-operated therapeutic communications. Table 1 summarizes previous review papers.
Summary of Reviews and Analyses of Personalized, Targeted, and Tailored Interventions
Theoretic Models
Patients are increasingly involved in managing their health care,2 and health care providers are challenged to motivate, educate, and help people adhere to healthy behaviors and medication regimens in the ambulatory setting.3 Understanding why people behave the way they do and identifying the factors underlying behavioral change help in the development and evaluation of effective health behavior interventions.4 Although a review of theories is outside the scope of this paper, we mention the four cognitive-behavioral models used most frequently in the studies reviewed.
Cognitive-behavioral theories focus on the individual level and use two key concepts—behavior (as mediated through cognitions) and knowledge (which is necessary but not sufficient to produce behavior change). These theories focus on intrapersonal factors such as an individual's knowledge, belifefs, motivation, attitudes, developmental history, experience, skills, self-concept, and behavior. Models using an intrapersonal approach are the stages of change, or transtheoretic, model (TM), the health belief model (HBM), and the theory of reasoned action/theory of planned behavior (TRA).5 6 The TM is concerned with an individual's readiness to change.6 7 The HBM focuses on an individual's perception of the threat of a health problem.4 The TRA focuses on an individual's intention to perform a behavior.6 Social-cognitive theory incorporates intrapersonal and interpersonal factors; as in the HBM and TRA, the benefits of a behavior must outweigh the costs; also, a person must have a sense of self-efficacy or personal agency about the behavior.8 9 Personal empowerment, an individual's ability to cope with situations and perceived sense of control over them, is emphasized.8 10 Table 2 summarizes the concepts of each theory.
Models and Concepts for Health Behavior Change
An appropriate theoretic framework applied to development of health behavior messages can greatly enhance a patient's motivation to comply with an intervention.5 Further enhancement can be achieved using patient characteristics in conjunction with computer production capabilities to approximate a face-to-face clinical encounter.6
Another enhancement feature may be achieved by using mobile devices rather than delivery methods that tether ambulatory patients to a computer, telephone, or mailbox. Mobile devices such as cell phones or pagers are particularly suitable for outpatient interventions, since patients can carry them easily. These devices have received considerable popular media and commercial attention,11 12 13 so we made an effort to find papers that described their use. The price of mobile communication devices has dropped dramatically in the last decade, so the increasing power and decreasing cost of communication may provide opportunities for therapeutic interventions that were not feasible before. In addition, their portability and convenience seem to create an attachment or synergy between the user and the device, which can bond the user to the intervention protocol.14
Methods
The literature that describes this area of investigation is not indexed in a single database. We therefore designed a search strategy that involved searching across multiple databases using both free text and appropriate specialized terminology.
Data Sources
Databases searched included MEDLINE (1966–99), HealthSTAR (1981–99), cinahl (1982–99), Current Contents (1997–99), embase (1990-1999), inspec (1969–1999), PsycINFO (1967–1999), and Sociological Abstracts (1986–1999). We also searched the Cochrane Collaboration and Web of Science (Science Citation Index Expanded and Social Sciences Index) databases. To account for gray literature, we searched CRISP (Computer Retrieval of Information on Scientific Projects) and Dissertation Abstracts, contacted authors, and conducted targeted Internet searches. We also searched LEXIS-NEXIS for more popular literature on this subject.
Searches were limited to publications in English. A summary of key terms and phrases is given in Table 3.
Key Words Used in Literature Review
Study Selection
Eligibility for inclusion in the final set included:
-
Controlled clinical trials and quasi-experimental studies with some evidence of instrument reliability and validity
-
At least one patient-interactive feedback, reminder, or educational intervention intended to influence or improve a stated health behavior
-
An association between one intervention variable and a health behavior
Eligible trials were evaluated using the rating system described in Table 4. Ratings were based on recommendations from the literature.15 16 Articles received a score from 1 to 10; sampling and randomization aspects and presence of a control group were weighted most heavily (totaling 7 points). The minimum score was set at 5 for inclusion.
Rating System
The initial cross-database search yielded 1,404 publications; this was reduced to 519 after elimination of duplicates. Review of the title and abstract of each publication yielded 97 publications potentially meeting eligibility criteria. After review of these articles, 49 publications were eliminated because they did not meet the eligibility criteria (the primary cause was a focus on physician reminders). Manual searches of the bibliographies of remaining articles, reviews, key journals in the appropriate fields, and key individuals yielded another 6 articles, for a total of 55; multiple reports of studies were collapsed to yield a final total of 46 studies in this review. Nine of these are feasibility or quasi-experimental studies included because they describe promising approaches. Figure 1 illustrates our selection process.
Search and selection process.
Data Extraction and Definitions
Each item was scored using the rating system described in Table 1. Items were classified by intervention type, delivery device, and use of synchronous vs. asynchronous interaction.
We defined three intervention types according to features accepted in the literature: personalized, targeted, and tailored. Personalized messages have the person's name on the information he or she receives. The message content is not adapted to the individual's diagnostic, behavioral or motivational characteristics.17 Personalized intervention studies were eliminated unless a higher level intervention (i.e., targeted or tailored, or both) was also a condition in the study.
Targeted message content is customized to reach a specific subgroup of the general population, based on the principles of market segmentation. Content is customized to “target” broad psychographic (i.e., activities, interests, and opinions) and sociodemographic groups. Targeted interventions do not account for personal differences in intervention needs among individuals in the target population, but they may be personalized.18
Tailored interventions are messages or a series of messages based on a specific individual's characteristics, as determined through historical records, replies to questions, or replies to previous messages. Tailored messages are generally based on published theoretic models, and message content is specific to one individual at one point in time. One of the goals of a tailored intervention is for patients to perceive the information as applying only to them.17 19 20 An example of a tailored intervention is delivering messages or information contingent on a patient's “stage of change,” a model postulating that patients will respond to and better remember messages presented on cue. For instance, patients who have just quit smoking will respond better to messages pitched to the “action” phase than the “precontemplative” phase.21 22 The actual messages are picked from a large pool of potential responses either manually by a therapist or through a largely automated process designed by a therapist.
The primary distinction between targeted and tailored interventions is that tailoring adapts content or the way content is presented according to the needs of the individual. Material is not fixed and feedback is based on individual, not subgroup, characteristics. Devices used for feedback range from the most simplistic, such as tailored letters, to expert systems incorporating behavior change models into an interactive messaging system.
Interventions can be distinguished along a continuum, from generic (or “one-size-fits-all”) to highly individualized, tailored approaches as seen in Figure 2. Confusion between targeted and tailored interventions is compounded by researchers' inconsistent use of the terminology.22
Intervention types continuum.
We grouped intervention delivery devices into categories adapted from Balas et al.23: 1) mobile communication systems (use of a pager, mobile telephone, or other wireless system for delivery), 2) computerized communication systems (use of a computer, modem, touch-sensitive screen, or other interfacing equipment for delivery), 3) automated telephone communication (usually computer-generated messages using a regular telephone line and telephone), and 4) print communication (use of a letter, bulletin, fax transmission, newsletter, postcard, or manual delivery).
We also distinguished between synchronous and asynchronous communication. Synchronous communication is like communication by telephone—dialog occurs in real time. Asynchronous communication is like e-mail—two parties carry on a dialog by leaving messages, but do not usually communicate in real time.
Results
Of the 46 studies meeting our inclusion criteria, 9 received scores below 5. These studies were excluded from analysis but are included in Table 5 because they illustrate promising approaches and merit discussion.
Summary of Targeted and Tailored Interventions, Categorized by Delivery Device
Lack of homogeneity among the remaining 37 studies precluded pooling of data; our findings, therefore, form a descriptive literature review. Studies varied by recruitment method, subject characteristics, study design, time frame, setting, measurement of effects, and health behavior focus. Many studies reported multiple outcomes; several used targeted or tailored interventions in conjunction with personalized or generic interventions; some used more than one targeted or tailored intervention.
Of 37 studies, 33 (89.2 percent) reported improved outcomes and 20 of these (60.6 percent) were statistically significant. Fourteen studies used targeted interventions; 23 used tailoring. Eleven of the targeted intervention studies (78.6 percent) reported improved outcomes; 6 of these (54.5 percent) were statistically significant. Of the 23 tailored intervention studies, 22 (95.7 percent) reported improved outcomes; 15 of these (68.2 percent) were statistically significant. Table 5 lists the studies by delivery device.
No Intervention Benefits
Four studies50 66 67,68 did not report statistically significant or improved outcomes. Lack of effect was explained by use of a limited, non-intensive one-time feedback with no inclusion of psychosocial factors66 and use of similar messages for both control and intervention protocols.68
Use of Models
Only 23 studies (62.2 percent) stated use of a theory to guide the health behavior intervention: 19 were print communications, 4 were telephone.
Comment on Emerging Technologies
Clinicians' use of pagers, personal digital assistants, PalmPilots, and laptop computers as portable information resource devices is a subject of numerous studies.69 Just as clinicians have found that these devices provide a greater sense of control, mastery, and personal empowerment in the work setting, perhaps patients may also find such devices advantageous when managing their treatment regimens in the outpatient setting. Portability of “always ready” devices in combination with the messaging interventions can create a synergistic feedback loop between patient and device as evidenced by Milch's finding64 that “several of the patients allowed that the pager became a trusted friend” and Dunbar's report65 of high patient engagement with a pager system.
Mobile systems may have clear advantages over computer, telephone, or print communication systems for delivery of tailored health behavior interventions, because they offer the benefits of constancy—”anytime, anywhere” messaging and communication capability14; physical freedom—because the system is wireless and mobile, the patient is not restricted to one physical environment to receive messages14; privacy—messages can be modified so that others cannot observe or interact with them14; and temporal flexibility—users can interact with the content when available or postpone interaction if desired.24
Yet a portable system is not without limitations and disadvantages. Potential problems include acceptance—studies reporting general acceptance of computerized and mobile delivery systems have had small numbers of subjects, so we do not know whether a selection bias or novelty variable is involved in this outcome; intrusiveness—the “anytime, anywhere” feature may be too intrusive for long-term use; and economic consequences—we do not know how a portable system may economically affect a patient, a practitioner, or the health care system. Only 13 studies in this review* considered cost as a factor of a system's feasibility or success.
Conclusion
Our review indicates that many studies demonstrate the effectiveness of “clinical encounters in absentia,” but few good studies incorporate leading edge communication technologies. The studies reviewed here represent the best available evidence to date of the effectiveness of targeted and tailored health behavior interventions across health behaviors.
An overwhelming majority (91.3 percent) of tailored intervention studies reported improved outcomes, as did 92.9 percent of those studies that used targeting; however, little research compared tailored to targeted interventions. We therefore cannot conclude that tailoring is more effective than targeting. One notable exception is Prochaska et al.41 Using the stages of change model to characterize readiness to quit smoking, they compared two tailored interventions to a targeted condition and general smoking cessation materials. At the 18-month follow-up period, tailored interventions outperformed both targeted and generic conditions and were associated with higher prolonged abstinence rates than other conditions. This study offers evidence that tailored interventions can more positively influence health behavior change than targeted, personalized, or generic interventions, but more studies like this need to be conducted.
There has also been little research comparing different tailoring protocols to one another. One group of smoking cessation studies compared three types of tailored interventions matched to stage of change to controls,39 42 four types of tailored interventions to a nontailored intervention,38 and two types of tailored interventions to generic material and controls.36 Two patterns emerged: there was more forward stage transition in the tailored groups compared with controls, and multiple interventions were more effective than single tailored interventions. Studies of this kind must be conducted for health behaviors other than smoking.
It is notable that only those studies using print and telephone devices reported a theoretical basis for their methodology. Future studies need to identify which models are best suited to which health behaviors and whether certain delivery devices are more appropriate for different health behaviors. While “it is not inconceivable to view computer technology for health promotion and the delivery of services as a form of medical intervention with patient satisfaction, compliance, and improved health status as the goals,”3 we need to know to what extent such interventions can beneficially replace interpersonal health behavior recommendations. In addition, usability studies could elucidate the complex process of human interaction with technology—the interplay between interface design and human cognition. The current research shows that isolated paper, telephone, and computer-delivered communications can cause health-enhancing behavior change. The communication of these behavior change models over an integrated Internet linked array of delivery devices, pagers, cell phones, interactive television, and computers with portable devices is technically feasible today.
Friedman et al.45 have stated that “in the future there will be devices in general use that will incorporate features of the current telephone, television, video, and computer as well as wireless devices that people will be able to carry with them.” That future is currently possible through the use of tetherless devices such as pagers and personal digital assistants, but research on these devices currently lags far behind research on print, telephone, and computerized communications.
In 1997, Balas et al.70 predicted that, “in the future, application of distance technology may strengthen the continuity of care between patient and clinician by improving access and supporting the coordination of health care activities from a single source.”23 Technology has finally reached the point that health behavior models can be integrated with computer-generated interventions to provide consistent, continuous interactive ambulatory care. What is missing is comprehensive measurement of the effectiveness of these systems, which has the potential to not only inform the organization and delivery of health care but help move the science of medical informatics toward the goal of achieving the status of a “mature science.”
Acknowledgments
The authors thank Sherrilynne S. Fuller, PhD, and Rona L. Levy, PhD, for helpful comments in the preparation of the manuscript.
