This study measured compliance with evidence-based guidelines after clinical alerts sent to physicians, compared with compliance after alerts sent to both physicians and their patients.
Objective: Clinical messages alerting physicians to gaps in the care of specific patients have been shown to increase compliance with evidence-based guidelines. This study sought to measure any additional impact on compliance when alerting messages also were sent to patients.
Study Design: For alerts that were generated by computerized clinical rules applied to claims, compliance was determined by subsequent claims evidence (eg, that recommended tests were performed). Compliance was measured in the baseline year and the study year for 4 study group employers (combined membership >100,000) that chose to add patient messaging in the study year, and 28 similar control group employers (combined membership >700,000) that maintained physician messaging but did not add patient messaging.
Methods: The impact of patient messaging was assessed by comparing changes in compliance from baseline to study year in the 2 groups. Multiple logistic regression was used to control for differences between the groups. Because a given member or physician could receive multiple alerts, generalized estimating equations with clustering by patient and physician were used.
Results: Controlling for differences in age, sex, and the severity and types of clinical alerts between the study and control groups, the addition of patient messaging increased compliance by 12.5% (P <.001). This increase was primarily because of improved responses to alerts regarding the need for screening, diagnostic, and monitoring tests.
Conclusion: Supplementing clinical alerts to physicians with messages directly to their patients produced a statistically significant increase in compliance with the evidence-based guidelines underlying the alerts.
(Am J Manag Care. 2008;14(11):737-744)
To maximize the impact on clinical quality and patient safety when clinical alerts are sent to physicians, they should be accompanied by similar messages to their patients.
The overall increase was because of significantly improved compliance with guidelinesrecommending screening, diagnostic, and monitoring tests.Numerous studies have documented the slow dissemination of new medical knowledge1,2 and the failure of many patients to receive important evidence-based clinical services.3,4 Clinical alerts can accelerate the dissemination of new knowledge and increase the use of evidence-based services by providing information on important drugs, tests, or other services that appear to be missing from a patient’s treatment.5-7 However, when alerts are sent only to physicians, these improvements often do not reach optimal levels.8,9 Busy physicians are inundated with messages from health plans and care management programs. Because of this, and the fact that messages from some sources may be perceived as unreliable, physicians may pay little attention to any clinical alerts. If physicians do take action based on alerts, their patients may not understand the recommendations or their importance, or may fail to fill the prescriptions or obtain the recommended tests.
These problems suggest that supplementing alerts to physicians with notices to their patients might be beneficial—encouraging patients to follow their physicians’ advice or to remind their physicians about overlooked guidelines. Patients and health plan members increasingly want to play an active part in their own care.10 Yet few consumer decision support programs are designed to empower the consumer at a point in time when a potential problem of clinical quality or safety has been detected, and fewer still are integrated with systems of physician alerts. As Glasziou and Haynes pointed out, full implementation of improvements in medical care requires not just dissemination of abstract knowledge, but also application of that knowledge by physicians to individual patients and, in most cases, actions by the patients themselves.11 It is not enough that a physician knows that medication X is now the drug of choice for condition Y. The physician must recognize that medication X is appropriate for patient Z and must write a prescription, and patient Z must fill and adhere to that prescription. Clearly, enhanced knowledge diffusion in the medical community alone is not enough. Clinical alerts to physicians concerning gaps in the care of specific patients can provide a useful reinforcement by directly addressing the applicability of new knowledge to individual patients. However, getting all the way to our goal may require including the patient in the system.
We studied the impact of a patient- messaging program designed to address these needs. Several previous studies documented the value of clinical alerts to physcians,5-7,12 and this study did not reexamine that issue. Our focus was on the incremental impact of supplementing a physician clinical alert system with information sent directly to patients concerning possible gaps in evidence-based care for their condition, with the 2 messages coordinated to enable the patient and physician to collaborate in closing those gaps.
Compliance with clinical alerts was based on claims evidence that the recommended actions (eg, to perform needed tests, to discontinue contraindicated medications) were actually carried out (eg, receipt of pharmacy claims documenting that patients filled prescriptions). For alerts suggesting the addition of a drug, test, vaccination, or other service, success was defined as claims evidence with a service date within 270 days after the alert was generated. For alerts recommending discontinuance of a drug, success was defined as the absence of a refill prescription between 60 and 150 days after generation of the alert.
The study was necessarily limited to “measurable” alerts, for which successful resolution could be determined from claims data. This excluded alerts recommending the avoidance of ginkgo biloba, or other outcomes knowable only from patient self-reports. Also eliminated from the study were newly implemented or discontinued alerts that were not in place during both the baseline and study years, and a small number of alert types that are never messaged to patients (eg, those that concern very sensitive topics such as HIV) or never messaged to physicians (eg, influenza immunizations that often are obtained from alternative sources).
A strenuous attempt was made to select similar employer groups for inclusion in the study and control groups, matching their membership as of January 2006 (the first month of the study period) in terms of average age, percent female, average risk score, and prevalence of 4 chronic conditions: diabetes, asthma, heart disease, and cancer. All employers in both groups had the same health insurer, health benefit design, and disease management program, and the 2 groups were matched for mix of industry types, using Standard Industrial Classification codes. Despite these efforts, significant differences in demographic characteristics, distribution of alerts, and compliance did exist between the study and control groups during the baseline year (). Therefore, in analyzing the impact of member messaging, we used multiple logistic regression to control for these variables. Compliance rates varied widely among types of alerts. To control for any differential impacts that year-to-year shifts in the mix of alerts might have on compliance rates in our study and control groups, compliance rates for each type of alert were included as independent variables in the regression. For the most stable estimates, we used the aggregate compliance rates for our entire client book of business, rather than rates for just the study and control group employers.
Members of the study and control groups who triggered alerts in the baseline year generated an average of 1.6 alerts each. Compliance with several alerts sent to the same patient (or the same physician) could not be considered independent. Therefore, our analyses used generalized estimating equations—the preferred method when analyzing correlated binary data—with clustering by both patient and physician.
Supplementing clinical alerts to physicians with messages directly to their patients produced a statistically significant increase in compliance with the evidence-based guidelines underlying the alerts. The overall increase seems to be due primarily to improved responses to advice regarding screening, diagnostic, and monitoring tests—advice that often is important but not temporally urgent. The improvement in compliance did not vary by patient age or sex.
Compliance with alerts is not the same as overall compliance with clinical guidelines. Alerts are issued only in those cases where evidence-based guidelines have not already been followed “spontaneously.” Apparent noncompliance can occur for clinically valid reasons, such as allergy to the recommended medication and other contraindications, which may be known to the treating physician but not recorded in claims data. A clinical guideline recommending anticoagulation therapy, for instance, is likely to be correctly ignored by a physician who knows that the patient falls frequently. However, noncompliance also may occur because of physician or patient lack of knowledge, understanding, or motivation. Alerts should not produce compliance in situations where it is not clinically advisable, and it is probably not reasonable to expect even an ideal system of alerts to totally overcome the inappropriate barriers to compliance. It should be stressed that our study measured only apparent noncompliance because patients may receive therapies not recorded in claims data (eg, medication samples).
Many factors can affect rates of compliance with evidence- based medical guidelines. Chief among them is the influence of medical journals, direct-to-consumer advertising, and other media that disseminate information to physicians and the general public. The use of a large control group, in which patients and their physicians would presumably be exposed to the same information environment as those in the study group, is the most basic form of control for this influence. Matching study and control employers in terms of their health insurance plan designs and the type of disease management program in which they participated (telephonic nurse counseling for chronic conditions) controlled for another group of factors known to affect compliance.
Compliance rates for some clinical alert outcome types were observed to decrease from the baseline year to the study year, mostly in the control group and markedly for either/ or alerts. Compliance rates for specific alerts vary widely, especially within a heterogeneous category like either/or alerts. If an alert with a low compliance rate begins to be issued more frequently—or an alert with a high compliance rate begins to be issued less frequently—it can depress the average level of compliance for an entire alert category. The latter appears to have occurred in the control group in 2006 for alerts that recommended either screening women more than 65 years of age for low bone density or treating them to prevent osteoporosis.
There are at least 3 possible mechanisms behind the increase in compliance with evidence-based guidelines when patient messages were added. It may be that patient messages served to remind patients and to reinforce the instructions that their physicians have given them, so that they were more likely to follow these instructions. It is also possible that patients, armed with the messages they receive from the system, reinforced the clinical alerts that their physicians received, thereby making it more likely that the physicians would write the prescriptions and order the tests in question. A third explanation is that a physician alert was directed to the incorrect physician for that patient or for that aspect of the patient’s care. The alerted patient, however, took the message to the correct caregiver. Our finding that the addition of patient messaging appeared to have its greatest impact on compliance with do-a-test alerts suggests that patient messages exerted most of their influence on the behavior of patients themselves, increasing the number who complied with physicians’ orders that involved time-consuming or unpleasant actions (eg, going to radiologists or laboratories for the performance of tests). Lack of a similar effect for patient alerts related to adding or stopping medications suggests that significant effort by patients was less of a barrier to compliance with these types of recommendations, which are largely under the control of their physicians, and that patients urging their physicians to follow guidelines may not be an important factor.
Interestingly, a previous study of this same program measured the impact of physician messaging and found 24% compliance with clinical alerts advising physicians to add a drug.12 The current study found remarkably similar levels of compliance with add-a-drug alerts in both the baseline and study years, in both the control and study groups, ranging from 23.8% to 27.4%. This finding would seem to lend support to the findings of the earlier study.
A clinically sound, evidence-based system for detecting possible gaps in care and bringing them to the attention of both patients and their physicians in a timely and constructive manner would benefit all segments of the population.14 As stated in Crossing the Quality Chasm,1 “tens of thousands of Americans die each year from errors in their care, and hundreds of thousands suffer or barely escape from nonfatal injuries that a truly high-quality care system would largely prevent.…”1,pg2 “In the area of effectiveness, there is considerable evidence that automated reminder systems improve compliance with clinical practice guidelines.”1,pg164
Such a system can be developed by adding patient messaging to an existing program of clinically advanced physician alerts, as demonstrated in this study, or by adding physician alerts to a system that began with patient reminders. Whichever approach is taken, the result should be one in which the 2 sets of messages are coordinated to reinforce each other and to strengthen the patient–physician relationship.
Author Affiliations: From the Division of Outcomes Research (SNR, AAW), and the Division of Health Informatics (TLS, IAJ), ActiveHealth Management, New York, NY.
Funding Source: The authors received no external support.
Author Disclosure: The authors are employees of ActiveHealth Management, a company that provides the clinical alert services evaluated in this study. Drs Rosenberg and Juster report owning stock in Aetna, the parent company of ActiveHealth Management.
Authorship Information: Concept and design (SNR, TLS, IAJ); acquisition of data (SNR, AAW); analysis and interpretation of data (SNR, TLS, AAW, IAJ); drafting of the manuscript (SNR, IAJ); critical revision of the manuscript for important intellectual content (SNR, TLS, AAW, IAJ); statistical analysis (TLS, AAW); administrative, technical, or logistic support (SNR); and supervision (SNR).
Address correspondence to: Stephen N. Rosenberg, MD, MPH, Division of Outcomes Research, ActiveHealth Management, 1333 Broadway, 4th Fl, New York, NY 10018. E-mail: email@example.com.
2. Cochrane LJ, Olson CA, Murray S, Dupuis M, Tooman T, Hayes S. Gaps between knowing and doing: understanding and assessing the barriers to optional health care. J Contin Educ Health Prof. 2007;27(2):94-102.
4. Stuart B. How Disease Burden Influences Medication Patterns for Medicare Beneficiaries: Implications for Policy. New York, NY: Commonwealth Fund; 2008.
6. McCowan C, Neville RG, Ricketts IW, Warner FC, Hoskins G, Thomas GE. Lessons from a randomized controlled trial designed to evaluate computer decision support software to improve the management of asthma. Med Inform. 2001;26(3);191-201.
8. Smith DH, Kramer JM, Perrin N, et al. A randomized trial of directto- patient communication to enhance adherence to b-blocker therapy following myocardial infarction. Arch Intern Med. 2008;168(5):477-483.
10. Kaplan S, Greenfield S. The patientâ€™s role in reducing disparities. Ann Intern Med. 2004;141(3):222-223.
12. Javitt JC, Steinberg G, Locke T, et al. Using a claims data-based sentinel system to improve compliance with clinical guidelines: results of a randomized prospective study. Am J Manag Care. 2005;11(2):93-102.
14. Bluman B, Jarvis-Selinger S, Hotz S. Guidelines implementation: transforming physician/patient interaction through stages of change. J Contin Educ Health Prof. 2007;27(3):188-189.