Search tips
Search criteria 


Logo of tmjMary Ann Liebert, Inc.Mary Ann Liebert, Inc.JournalsSearchAlerts
Telemedicine Journal and e-Health
Telemed J E Health. 2009 October; 15(8): 770–776.
PMCID: PMC2998278

A Study of Medication-Taking and Unobtrusive, Intelligent Reminding

Tamara L. Hayes, Ph.D.,corresponding author1 Kofi Cobbinah, B.S.,4 Terry Dishongh, Ph.D.,4 Jeffrey A. Kaye, M.D.,2 Janna Kimel, B.A.,4 Michael Labhard, M.D.,4 Todd Leen, Ph.D.,3 Jay Lundell, Ph.D.,4 Umut Ozertem, Ph.D.,3 Misha Pavel, Ph.D.,1 Matthai Philipose, Ph.D.,4 Kevin Rhodes, B.S.,4 and Sengul Vurgun, M.S.4


Poor medication adherence is one of the major causes of illness and of treatment failure in the United States. The objective of this study was to conduct an initial evaluation of a context-aware reminder system, which generated reminders at an opportune time to take the medication. Ten participants aged 65 or older, living alone and managing their own medications, participated in the study. Participants took a low-dose vitamin C tablet twice daily at times that they specified. Participants were considered adherent if they took the vitamin within 90 minutes (before or after) of the prescribed time. Adherence and activity in the home was measured using a system of sensors, including an instrumented pillbox. There were three phases of the study: baseline, in which there was no prompting; time-based, in which there was prompting at the prescribed times for pill-taking; and context-aware, in which participants were only prompted if they forgot to take their pills and were likely able to take their pills. The context-based prompting resulted in significantly better adherence (92.3%) as compared to time-based (73.5%) or no prompting (68.1%) conditions (p < 0.0002, χ2 = 17.0). In addition, subjects had better adherence in the morning than in the evening. We have shown in this study that a system that generates reminders at an opportune time to take the medication significantly improves adherence. This study indicates that context-aware prompting may provide improved adherence over standard time-based reminders.

Key words: home health telemonitoring, telehealth, information management


Poor medication adherence is one of the major causes of illness and of treatment failure in the United States (U.S.). Bedell et al.1 found that 76% of prescription and nonprescription medicines are taken incorrectly. This is particularly true for the aging. More than 75% of people aged 65 and older take prescription medication, and on average they take three or more medications a day.2,3 Unfortunately, more than 50% of these individuals are nonadherent to their medication regimen,4,5 which can lead to increased hospitalization and drug side-effects.6 With more than 3 billion prescriptions worth $203 billion dispensed in the U.S. annually,7 it is clear that tools to help people take their medications properly can be of tremendous benefit. Furthermore, the benefit of such tools extends beyond the management of medication adherence in patient populations. Clinical trials to assess the safety and efficacy of new drugs necessarily rely on proper medication adherence by study participants to obtain accurate data.8,9 Most interventions to improve medication compliance are complex and require frequent contact between a patient and a nurse or caregiver.1013 Thus, methods to both accurately assess medication adherence and to encourage optimal adherence are of considerable interest.

Existing methods of tracking medication adherence suffer from a number of drawbacks. Pill counts, the mostly commonly used method, overestimate adherence.1416 So does self-report of adherence.14,17,18 The widely used Medication Event Monitoring System (MEMS, Aardex Ltd.) provides excellent information about adherence, but has several shortcomings. The MEMS cap is difficult to open for arthritic hands,1921 and it does not accommodate the use of pill boxes for sorting medications into daily doses,22 as are commonly used by the elderly23 or when taking multiple drugs.22 MEMS also does not report adherence in real-time, so rapid intervention cannot take place.

Reminders and aids to adherence vary in their usefulness.12,2426 In general, multimodal methods that incorporate behavioral feedback appear to be the most effective at improving adherence outcomes.27 However, most successful interventions appear to require active intervention by a human monitor, and in general, simple time-based prompting is not as effective as reminders that are behaviorally integrated into a person's daily activities.23 With this in mind, we hypothesized that a system which generates reminders at an opportune time to take the medication will lead to improved adherence as compared to time-based reminders. To examine this hypothesis, we have developed a medication tracking and reminding system that allows both measurement of adherence, and real-time, context-aware reminders.28 The objective of the present study was to evaluate whether or not such context-aware reminders may improve adherence as compared to time-based reminders. It should be noted that the research system used in this study is more complex than would ultimately be necessary for a context-aware medication reminder system, since it was also used to develop the inference algorithms needed to determine when prompting is appropriate. There are technical challenges inherent in deploying such a research system to a large number of homes that we have been addressing in related studies. This report summarizes the results of our initial evaluation in a small group of community-dwelling elders.

Materials and Methods


Ten participants aged 65 or older were selected from a pool of participants in an earlier study of medication adherence in which they were considered “poor adherents” (missing more than 20% of prescribed doses during a twice-daily vitamin regimen over a 5-week period).29 All participants lived independently and alone in one of two retirement communities and were currently managing their own medication regimen on a daily basis. All participants provided written informed consent (Oregon Health & Sciences University Institutional Review Board # 1682). Participants were functionally independent, and were screened for dementing illness with the Mini-Mental State Examination30 (exclusionary score <24) and the Clinical Dementia Rating scale31 (exclusionary score >0).


Age and years of schooling were collected for each participant. Cognitive function was assessed using the Alzheimer's Disease Assessment Scale–Cognitive Subtest (ADAS-Cog).32,33 The ADAS-Cog has been used to differentiate the stages of Alzheimer's disease from mild to severe,34 and there exist published norms for this test in a range of ages of elderly adults.33 In addition, we also collected delayed word-list recall scores,35 scores on both parts A and B of the Trail Making Test,36 scores on the Geriatric Depression Scale,37 and scores on the Cumulative Illness Rating Scale.38

Medication Monitoring and Reminding System

In order to provide context-aware prompting to participants when they missed taking their medications on time, we created a medication monitoring and reminding system that unobtrusively collected data about the participants' activities and medication adherence, and used this information to prompt at appropriate times.28 Table 1 shows the components of the system and their purpose. The prompting was controlled by an inference engine, which used dynamic Bayesian inference39,40 to “learn” an individual's activities from the sensor data and then to make a decision to prompt based on those activities.

Table 1.
Equipment Used in Each Home to Enable Context-Aware Prompting


Participants were asked to take a low-dose (250 mg) vitamin C tablet twice daily at 12-hour intervals (e.g., 7 am and 7 pm); the times were selected by the subject during the initial interview. Participants were considered adherent if they took the vitamin within 90 minutes (before or after) of the prescribed time. We used the medication tracking and reminding system to assess how different forms of prompting (no prompting, time-based prompting, and context-aware prompting) affected adherence in this group, using a three-period repeated-measures design. During the no-prompting phase (10.7 ± 4.4 weeks), participants received no reminders to take their vitamins. During time-based prompting (10.1 ± 3.3 weeks), participants were reminded to take their vitamin by the auditory beep and visual alarm on the MedTracker device, which would prompt for 10 seconds every minute for 3 minutes at the specified dose times. During context-aware prompting (8.4 ± 4.2 weeks), prompting occurred only when the system inferred that participants were likely to miss taking their vitamin, and at a time when the participant was likely able to take the pill. The generation of the prompting events (signals), consistent with the inference, was governed by the following set of rules:

  1. Prompt at the closest location to the participant, using either the MedTracker itself or the closest activity beacon.
  2. Never prompt outside the adherence window (90 minutes before and after the time to take the vitamin).
  3. Don't prompt if the vitamin has already been taken within the current window.
  4. Don't prompt if the participant is not home. Resume if the participant returns home before the window ends.
  5. Don't prompt if the participant is in bed. Wait until they are out of bed to prompt.
  6. Don't prompt if participant is on the phone. Wait until they are off the phone to prompt.
  7. Don't prompt before the time the user usually takes the pill (estimated from their baseline data). The only exception to this is if the participant appears to be leaving the home prior to their regular pill-taking time.
  8. If it is less than 20 minutes left until the adherence window expires, start prompting, disregarding all other rules (except 2–4).

In the case of the context-aware prompting, participants were prompted both by the visual and auditory alarms on the MedTracker, and by the flashing and audio message of the beacon (Table 1).


We defined “adherent” as taking the vitamin within a 3-hour window around the selected time–90 minutes before the time and 90 minutes after the time. Adherence to regimen was measured as the percentage of pills taken within this window. Consistent with past studies of medication taking, an adherence of 80% or greater was considered good adherence.4143

The effect of prompting strategy on adherence was assessed using a mixed model design, with subjects as a random factor and prompting strategy as a fixed factor, with the weeks of monitoring as a repeated measure within each prompting strategy.

We constructed a one-way analysis of variance (ANOVA) based on nested beta-binomial residuals for the adherence data. This analysis is more appropriate for binomial data than the usual Gaussian-based ANOVA (particularly for very high or low adherence rates), and accounts for the variation between subjects (random effect) with a beta distribution on the underlying adherence rate.44 We constructed three nested beta-binomial models for this analysis. The simplest used a single distribution to model all the data (two parameters), and hence assumed that the adherence rates were independent of the prompting strategy. The second model allowed the mean population rate to depend on the prompting strategy but constrained the variances, resulting in four model parameters. The third model removed the constraint on the variances, resulting in six free parameters. We fit the models by maximum likelihood and compared them using a likelihood ratio test. We estimated the 95% confidence intervals from the distribution of means computed on 2,000 bootstrap replicates from the original data, each fit by maximum likelihood to the model selected by the ANOVA. Multiple comparisons were made using the Bonferroni criterion.


Ten subjects (mean age 82.7 ± 6.4, 9 females, 1 male, mean education 16.3 ± 2.2 years) participated in the study. Cognitive and health measures for these subjects are shown in Table 2. Two subjects dropped out of the study prior to the introduction of the first prompting intervention (1 due to illness, 1 chose not to continue). Another subject was removed from the study due to violation of the protocol. In other cases, only partial data were available due to withdrawal after the time-based prompting (2 subjects). Thus, complete data were available for only 7 subjects. One final subject had only partial data during the baseline phase due to technical issues. The data from that subject were included in the analyses below, since it allowed us to look at differences between time-based and context-based prompting. There were no differences in any of the neurocognitive measures between the subjects used for analysis and the complete group (Table 2).

Table 2.
Group Characteristics (Mean ± Standard Deviation)

The best fit to the adherence data resulted from the second ANOVA model, in which the population mean adherence rates depended on the prompting strategy but the variances were constrained. For most subjects, their adherence was quite variable week to week. The mean adherence rates (estimated in the ANOVA) and their 95% confidence intervals was 68.1% [57.5–80.5] for the baseline condition. During this period, only 1 subject achieved good adherence (>80%). Although subjects did better overall during the time-based prompting phase, only 2 subjects obtained an average weekly adherence of over 80%. The mean adherence rate was 73.5% [68.0–78.6] for time-based prompting. Notably, during the context-aware prompting phase, most subjects achieved excellent adherence, and the mean adherence was 92.3% [84.7–97.0].

The differences in the mean adherence rates for each prompting strategy were significant at p < 0.0002 (χ2 = 17.0). Figure 1 shows the estimated beta distributions corresponding to each of the three prompting conditions. Context-aware prompting had a clear advantage over the baseline and time-based prompting; not only is the mean adherence rate higher, but the variation is less than for the baseline or time-based prompting populations. There was also a small difference between subjects (F6,169 = 2.7, p < 0.02).

Fig. 1.
Estimated Beta distributions corresponding to each of the three prompting conditions.

The mean difference in adherence during context-based prompting as compared to the no-prompting phase was 24.1%, and between context-based and time-based was 14.4%. A box plot of overall, morning, and evening adherence (Fig. 2) reveals some interesting details of the participants' adherence during the study. In particular, evening adherence was consistently worse than adherence in the morning during all prompting phases; however, the difference between morning and afternoon adherence under context-based prompting was much less.

Fig. 2.
Box plot of weekly morning, evening, and overall adherence rates under each of the prompting conditions. Asterisks indicate the value of outliers.


We have shown in this study that adherence to a medication regimen can be improved through the introduction of electronic reminders, and in particular that a system that generates reminders at an opportune time to take the medication significantly improves adherence. With context-based prompting, reminders occurred only when the system inferred that participants were likely to miss taking their vitamin, and at a time when the participant was likely able to take the pill. Although simple time-based reminders were somewhat effective at improving our participant's adherence, the context-based system led to excellent compliance: 86% of participants showed good adherence during that phase of the study, as compared to only 17% during the no-prompting and time-based phases of the study. To our knowledge, this is the first study to demonstrate the efficacy of a context-based medication reminder system.

Many interventions have been proposed for improving medication adherence.12,13,24 Recently, a number of electronic devices that sound an alarm and dispense medication45 when a pill should be taken have become commercially available. Unfortunately, like a wrist-watch alarm, all of these devices require the user to be present to hear the alarm. More recently, Sterns and Mayhorn proposed a pill-dispensing Personal Digital Assistant (PDA), which led to improved adherence in a population of young adults during the third month of a 3-month field trial.46 Because subjects carried the PDA during the day, they were able to take the medication from the device when prompted. However, this approach has the drawback that the user must remember to carry the device, and in some populations, including the elderly, this may pose an unacceptable burden. The system used in our study did not require the user to carry any devices, but rather monitored their movements and medication taken unobtrusively and determined when prompting should be done.

Our study had some limitations. The sample size is this exploratory study was very small, and as a consequence we were unable to conduct the study as a cross-over study. Therefore, we cannot know for sure that adherence simply improved over time for some of our participants. However, subjects were followed for several months in each phase of the study, and we did not see trends within individual phases. Our finding of somewhat increased adherence during the time-based prompting phase is also consistent with other studies of time-based prompting. The small sample size also makes it difficult to draw general conclusions about the efficacy of our approach in a large population. Nonetheless, the results of the current study are extremely promising.

This study has important implications for elders living independently in the community. Medication mismanagement in the elderly is a significant health issue, both financially and in terms of increased hospitalization47 and drug side-effects.6 Many elders would prefer to remain in their homes, and yet there are few effective tools available for managing the large number of medications taken by this population. The results of the current study suggest that a context-aware medication reminding system could lead to improved medication adherence in this population. Further investigation in a larger study of community-dwelling elders is needed to confirm the potential of this approach.


The authors thank Intel Corporation for their support of this research. Dr. Kaye's effort on this research was supported by National Institutes of Health (NIH) grants #AG024059, #AG024978. Dr. Hayes' effort on this research was funded by Oregon Opportunity.

Disclosure Statement

This study was funded by Intel Corporation, a company that may have a commercial interest in the results of this research and technology. Kofi Cobbinah, Terry Dishongh, Janna Kimel, Michael Labhard, Jay Lundell, Umut Ozertem, Matthai Philipose, Kevin Rhodes, and Sengul Vurgun were Intel employees at the time of this study. In addition, Tamara Hayes has a significant financial interest in Intel Corporation. This potential conflict has been reviewed and managed by OHSU. Jeffrey Kaye and Misha Pavel do not have any competing financial interests in this study.


1. Bedell SE. Jabbour S. Goldberg R, et al. Discrepancies in the use of medications: Their extent and predictors in an outpatient practice. Arch Intern Med. 2000;160:2129–2134. [PubMed]
2. Helling DK. Lemke JH. Semla TP. Wallace RB. Lipson DP. Cornoni-Huntley J. Medication use characteristics in the elderly: The Iowa 65+ Rural Health Study. J Am Geriatr Soc. 1987;35:4–12. [PubMed]
3. Ostrom JR. Hammarlund ER. Christensen DB. Plein JB. Kethley AJ. Medication usage in an elderly population. Med Care. 1985;23:157–164. [PubMed]
4. Botelho RJ. Dudrak R., 2nd Home assessment of adherence to long-term medication in the elderly. J Fam Pract. 1992;35:61–65. [PubMed]
5. Kendrick R. Bayne JR. Compliance with prescribed medication by elderly patients. Can Med Assoc J. 1982;127:961–962. [PMC free article] [PubMed]
6. Col N. Fanale JE. Kronholm P. The role of medication noncompliance and adverse drug reactions in hospitalizations of the elderly. Arch Intern Med. 1990;150841 [PubMed]
7. NACDS. Community Pharmacy Results: National Association of Chain Drug Stores. 2003. [Aug 27;2009 ].
8. Gorkin L. Goldstein MG. Follick MJ. Lefrbvre RC. Strategies for enhancing adherence in clinical trials. In: Schumaker SA, editor; Schron EB, editor; Ockene JK, editor. The handbook of health behavior change. New York: Springer; 1990. pp. 361–375.
9. Dunbar-Jacob J. Burke LE. Puczynski S. Clinical assessment and management of adherence to medical regimens. In: Smith TW, editor. Managing chronic illness. Washington, DC: American Psychological Association; 1995. pp. 313–350.
10. Becker M. Theoretical models of adherence and strategies for improving adherence. In: Schumaker SA, editor; Schron EB, editor; Ockene JK, editor. The handbook of health behavior change. New York: Springer; 1990. pp. 5–43.
11. Rand CS. Issues in the measurement of adherence. In: Shumaker SA, editor; Schron EB, editor; Ockene JK, editor. The handbook of health behavior change. New York: Springer; 1990. pp. 102–110.
12. McDonald HP. Garg AX. Haynes RB. Interventions to enhance patient adherence to medication prescriptions: Scientific review. JAMA. 2002;288:2868–2879. [PubMed]
13. Haynes RB. McDonald H. Garg AX. Montague P. Interventions for helping patients to follow prescriptions for medications. Cochrane Database Syst Rev. 2002;2 CD000011. [PubMed]
14. Carroll CL. Feldman SR. Camacho FT. Manuel JC. Balkrishnan R. Adherence to topical therapy decreases during the course of an 8-week psoriasis clinical trial: Commonly used methods of measuring adherence to topical therapy overestimate actual use. J Am Acad Dermatol. 2004;51:212–216. [PubMed]
15. Cramer JA. Mattson RH. Prevey ML. Scheyer RD. Ouellette VL. How often is medication taken as prescribed? A novel assessment technique. JAMA. 1989;261:3273–3277. [PubMed]
16. Namkoong K. Farren CK. O'Connor PG. O'Malley SS. Measurement of compliance with naltrexone in the treatment of alcohol dependence: Research and clinical implications. J Clin Psychiatry. 1999;60:449–453. [PubMed]
17. Olivieri NF. Matsui D. Hermann C. Koren G. Compliance assessed by the Medication Event Monitoring System. Arch Dis Child. 1991;66:1399–1402. [PMC free article] [PubMed]
18. Shalansky SJ. Levy AR. Ignaszewski AP. Self-reported Morisky score for identifying nonadherence with cardiovascular medications. Ann Pharmacother. 2004;38:1363–1368. [PubMed]
19. Whyte LA. Medication cards for elderly people: A study. Nurs Stand. 1994;8:25–28. [PubMed]
20. Keram S. Williams ME. Quantifying the ease or difficulty older persons experience in opening medication containers. J Am Geriatr Soc. 1988;36:198–201. [PubMed]
21. Atkin PA. Finnegan TP. Ogle SJ. Shenfield GM. Functional ability of patients to manage medication packaging: A survey of geriatric inpatients. Age Ageing. 1994;23:113–116. [PubMed]
22. Wendel CS. Mohler MJ. Kroesen K. Ampel NM. Gifford AL. Coons SJ. Barriers to use of electronic adherence monitoring in an HIV clinic. Ann Pharmacother. 2001;35:1010–1015. [PubMed]
23. Branin JJ. The role of memory strategies in medication adherence among the elderly. Home Health Care Serv Q. 2001;20:1–16. [PubMed]
24. Haynes RB. McKibbon KA. Kanani R. Systematic review of randomised trials of interventions to assist patients to follow prescriptions for medications. Lancet. 1996;348:383–386. [PubMed]
25. Leirer VO. Morrow DG. Pariante GM. Sheikh JI. Elders' nonadherence, its assessment, and computer assisted instruction for medication recall training. J Am Geriatr Soc. 1988;36:877–884. [PubMed]
26. Fishkin K. Wang M. Intel Report IRS-TR-03-011. Seattle, WA: Intel Corporation; Oct, 2003. A flexible, low-overhead ubiquitous system for medication monitoring.
27. Kripalani S. Yao X. Haynes RB. Interventions to enhance medication adherence in chronic medical conditions: A systematic review. Arch Intern Med. 2007;167:540–550. [PubMed]
28. Lundell J. Hayes TL. Vergun S, et al. Continuous activity monitoring and intelligent contextual prompting to improve medication adherence. Paper presented at: 29th Annual International Conference of the IEEE Engineering in Medicine and Biology Society; Lyon, France. 2007. [PubMed]
29. Hayes TL. Hunt JM. Adami AM. Kaye J. An electronic pillbox for continuous monitoring of medication adherence. Paper presented at: 28th Annual International Conference of the IEEE Engineering in Medicine and Biology Society; Aug 30-September 3;2006 ; New York, NY. [PMC free article] [PubMed]
30. Folstein M. Folstein S. McHugh P. “Mini-mental state”: A practical method for grading the cognitive state of patients for the clinician. J Psychiatr Res. 1975;12:189–198. [PubMed]
31. Morris J. The clinical dementia rating (CDR): Current version and scoring rules. Neurology. 1993;43:2412–2414. [PubMed]
32. Zec RF. Landreth ES. Vicari SK, et al. Alzheimer disease assessment scale: Useful for both early detection and staging of dementia of the Alzheimer type. Alzheimer Dis Assoc Disord. 1992;6:89–102. [PubMed]
33. Graham DP. Cully JA. Snow AL. Massman P. Doody R. The Alzheimer's Disease Assessment Scale-Cognitive subscale: Normative data for older adult controls. Alzheimer Dis Assoc Disord. 2004;18:236–240. [PubMed]
34. Zec RF. Landreth ES. Vicari SK, et al. Alzheimer Disease Assessment Scale: A subtest analysis. Alzheimer Dis Assoc Disord. 1992;6:164–181. [PubMed]
35. Welsh K. Breitner J. Magruder-Habib K. Detection of dementia in the elderly using telephone screening of cognitive status. Neuropsych Neuropsychol Behav Neurol. 1993;6:103–110.
36. Drane DL. Yuspeh RL. Huthwaite JS. Klingler LK. Demographic characteristics and normative observations for derived-trail making test indices. Neuropsych Neuropsychol Behav Neurol. 200215:39–43. [PubMed]
37. Sheikh J. Yesavage J. Clinical gerontology: A guide to assessment and intervention. New York: The Haywood Press; 1986. Geriatric Depression Scale (GDS): Recent evidence and development of a shorter version; pp. 165–173.
38. Parmelee P. Thuras P. Katz I. Lawton M. Validation of the Cumulative Illness Rating Scale in a geriatric residential population. JAGS. 1995;43:130–137. [PubMed]
39. Vurgun S. Philipose M. Pavel M. A statistical reasoning system for medication prompting. In: Krumm J, editor; Abowd GD, editor; Seneviratne A, editor; Strang T, editor. UbiComp 2007: Ubiquitous Computing, 9th International Conference. Vol. 4717. Berlin: Springer-Verlag; 2007. pp. 1–18.
40. Fox V. Hightower J. Liao L. Schulz D. Borriello G. Bayesian filtering for location estimation. Pervasive Computing IEEE. 2003;2:24–33.
41. Ho PM. Magid DJ. Masoudi FA. McClure DL. Rumsfeld JS. Adherence to cardioprotective medications and mortality among patients with diabetes and ischemic heart disease. BMC Cardiovasc Disord. 2006;6:48. [PMC free article] [PubMed]
42. Ho PM. Rumsfeld JS. Masoudi FA, et al. Effect of medication nonadherence on hospitalization and mortality among patients with diabetes mellitus. Arch Intern Med. 2006;166:1836–1841. [PubMed]
43. Osterberg L. Blaschke T. Adherence to medication. N Engl J Med. 2005;353:487–497. [PubMed]
44. Crowder MJ. Beta-binomial Anova for proportions. Applied Statistics. 1978;27:34–37.
45. Szeto AYJ. Giles JAJ. Improving oral medication compliance with an electronic aid. IEEE Eng Med Biol. 1997;16:48–54. [PubMed]
46. Sterns AA. Mayhorn CB. Persuasive pillboxes: Improving medication adherence with personal digital assistants. Paper presented at: 1st International Conference on Persuasive Technology for Human Well-Being, PERSUASIVE 2006; May 18–19;2006 ; Eindhoven, Netherlands.
47. Yee JL. Hasson NK. Schreiber DH. Drug-related emergency department visits in an elderly veteran population. Ann Pharmacother. 2005;39:1990–1995. [PubMed]

Articles from Telemedicine Journal and e-Health are provided here courtesy of Mary Ann Liebert, Inc.