Effects of Physician Communication and Family Hardiness on Patient Medication Regimen Beliefs and Adherence

Abstract

This study aims to develop a model that describes how physician communication and family hardiness affect medication regimen beliefs and adherence for patients on regimens to control diabetes and hyperlipidemia (high cholesterol). Study participants (n=1356) completed self-report questionnaires on health beliefs. Pharmacy refill records from a health plan in the United States provided data on their medication adherence. Structural equation modeling was used to model variable relationships. A mediation analysis demonstrated that physician communication behaviors had a significant impact on the patients’ behavioral intention to adhere to their regimen and medication adherence when they were mediated by the patient's medication taking health beliefs related to perceived benefit of the regimen, impact of side effects, and cost of regimen. Conversely, family hardiness had no effect on medication-taking behaviors. To improve patient medication-taking behaviors, physician communication behaviors should be targeted. The study suggests the physician's initial regimen discussion is important to both regimen initiation and long-term adherence, and should emphasize the regimen's benefits and how to avoid common side effects. Also, establishing a follow-up physician-patient relationship can enhance regimen adherence and reduce the likelihood that a patient will stop taking the medication due to cost concerns. The research supports the important role the physician plays in health behavior maintenance. Future research should study the effect physicians have on other recurring health behaviors.

Introduction

Medication therapy is vital for treatment of chronic diseases, but non-adherence to medication is a primary reason for treatment failure in diseases such as heart disease and diabetes [1-3]. Untreated heart disease is the leading cause of mortality in the United States (U.S.) [4]. Diabetes is the leading cause of kidney disease, blindness, and lower limb amputation among adults, aged 20–74 [5]. For heart disease, statin medications reduce major vascular events by 20 percent [6]. Yet, just 40–60 percent of those on a statin regimen adhere to it sufficiently to experience the benefits of the treatment [7,8]. For diabetes, the clinical impact of many simple and effective diabetic therapies has been limited by poor adherence rates [9]. Despite the medical community's increased reliance on medical therapy to treat disease, medication adherence rates have remained relatively unchanged for 40 years [10-12]. As a result, non-adherent patients have poorer clinical outcomes and higher medical costs than medication-adherent patients [3,13].

Understanding and predicting medication adherence is complicated by the diverse set of factors affecting this behavior [14]. The roles of the patient's physician and family have received considerable attention in medication adherence research (15,16). Medication adherence interventions tend to take place in the physician's office or the patient's home because of the role these settings play in a patient's health care maintenance and daily living [17].

Two Social Factors of Medication Adherence

Clinician-patient communication is considered an important and modifiable determinant of patient adherence behaviors. The physician communication literature has found that positive physician communication patterns result in 1.64 times higher patient adherence [16]. Positive physician communication patterns are enhanced during the initial regimen discussion by using patient-centered counseling that proactively engages the patient [18]. When the patient believes that there is concordance between their preferences and the physician prescribed regimen, greater medication adherence rates result [19]. Podl et al. [20] also found that the physician's communication style is important during both the initial regimen discussion and follow-up appointments. These communications offer opportunities to sustain individual motivation, assess progress, provide feedback, and adjust behavior plans. Providers who develop “relationships” that foster continuity, trust, and accountability tend to have more adherent patients [21]. It is less clear, however, which aspects of the provider's communications patterns are most critical for maintaining medication adherence behaviors and how these patterns influence patient health beliefs [22]. A patient's health beliefs are considered to be an important antecedent to the intention to perform and actually engage in a health behavior [23,24]. Patient health beliefs are directly related to medication adherence behavior [25]. Investigating the causal chain that occurs between provider communication patterns, the development of patient health beliefs, and medication adherence could provide insights into how patients arrive at medication adherence decisions [26]. In practice, understanding which of these pathways are most relevant to medication adherence behavior could guide physicians in how to design meaningful communication patterns to enhance medication adherence.

The effectiveness of physician-patient communication may not be an isolated activity and could be affected by the environment patients will return to following their physician visits. Family environment does impact medication adherence [27]. Medication adherence is 1.74 times higher in cohesive families and 1.53 times lower in families with conflict [15]. Several theories suggest that family resilience or hardiness can help individual members cope with stressors in the environment, leading to better chronic disease outcomes [15]. Additionally, a family with high resilience can use an adverse health episode to strengthen the family and actually improve the health of family members [28]. How these traits influence medication adherence beliefs and behaviors could affect how patients process and act on medical information and advice provided by physicians.

Study Model and Hypotheses

The aim of the study was to a) model how physician communication patterns and family hardiness affect behavioral beliefs, and b) predict six-month medication adherence for patients on an existing regimen. The theoretical model was based upon the Theory of Planned Behavior (TPB) [29]. The TPB projects that all health beliefs will influence behavioral intent, which, in turn, influences the actual behavior. In contrast to its predecessor, the Theory of Reasoned Action, TPB includes control beliefs that describe how individuals can control behavior [30]. Control beliefs are projected to influence both behavioral intent and actual behavior. We hypothesize that a model with physician communication and family hardiness external variables will influence patient health beliefs, which in turn will influence both the behavioral intent to adhere to the medication and their medication adherence behavior (Figure 1). The only control belief in the model is “Money for Regimen,” which is projected to influence both behavioral intent and medication adherence.

Figure 1.

A hypothesized Physician-Patient Communication Model of Medication Adherence – Goes About Here

Methods

2.1 Design

This longitudinal study assessed the relationship of selected external (exogenous) factors and health beliefs on medication adherence over a six-month period, from April 1 to September 30, 2003. The external factors and health beliefs were assessed using a structured questionnaire mailed in February 2003 to 6,402 adult health plan (United Healthcare) members from a national sample drawn specifically for this study. The representativeness of the survey sample was assessed by comparing characteristics of the respondents against the characteristics of the non-respondents.

2.2 Sample, sampling, and data collection

Individuals identified in the sample selection met the eligibility criteria of: a claim submitted for HMG-CoA reductase inhibitors (statins) or oral diabetic agents; continuous enrollment in the health plan for six months extending through the last month used for sample selection; age greater than 21; and enrollment in a family plan. HMG CoA reductase inhibitors (statins) for hyperlipidemia and oral diabetic agents (glipizide, glimepiride, glyburide, metformin, and rosiglitazone) were selected as the target study medications because of their high incidence in managing chronic disease, impact on health, and because they require long-term daily use for chronic disease management. In addition, all medications were used solely to treat a specific disease at the time of the trial. This allowed for the differentiation between medication switching (to another drug) and non-adherence to a medication regimen. It also permitted the researchers to assign the hyperlidemia and diabetes diagnoses.

Eligible participants were contacted by mail and asked to complete the University of Wisconsin Institutional Review Board (IRB) approved self-report survey instrument. The instrument collected information about their beliefs and intentions regarding adherence to the targeted medication during the two months prior to measuring the study's medication adherence variable.

Three thousand two hundred and thirty-two returned surveys were received from the sample of 6,402 for a response rate of 50.4%. Of the returned surveys, 339 subjects left the health plan during the period measuring medication adherence, and 169 either had their regimen changed by their physician or disclosed that they had completed the survey for a family member. These subjects were removed from the dataset. In addition, the dataset contained 80 missing values (missing % = 2.48%). A correlation analysis was used to test if the missing data was missing completely at random (MCAR), missing at random (MAR), or not missing at random (NMAR). If the MCAR condition was not met, multiple imputation of missing data points would have been performed [31]. Running a correlation of missing values to discover if “missingness” in one variable is present in another tests MCAR [31]. The correlation analysis of missing values found correlation among no variable in the study. Due to the low number of missing values and no correlations of variables being found, list wise deletion of missing values was employed. Health plan disenrollment, prescription discontinuation, survey completion errors, or the presence of missing data resulted in a final dataset of 2,644 subjects (from the 3,232 returned surveys).

2.3 Survey Instrument

The patients’ perceptions of the medication external variables and adherence beliefs were assessed through a written questionnaire. The questionnaire included factors projected to have greatest impact on long-term medication use. These factors were derived by an expert panel using an evidence-based solicitation model from the decision science field [32]. The 15-member panel included representatives from medicine, nursing, psychiatry, pharmacy, and public health who each had published multiple peer-reviewed publications on medication adherence. The panel applied a conjoint modeling decision science approach that arrived at the set of external factors and health beliefs. The approach consisted of the following steps: 1) Elicit an initial set of factors through expert interviews and an extensive literature review; 2) Finalize the list of factors and assign levels of impact for each one (e.g., no side effects, minimal side effects, etc.); 3) Validate that the measures and levels appear in existing research; 4) Develop hypothetical profiles that include different combinations of all factors; 5) Score the profiles, and 6) Select factors thought to have the greatest relationship with medication adherence by choosing external variables and health beliefs based on profile results with the highest scores.

2.4 Measures

The patient questionnaire was populated with the full set of expert-derived measures. The dataset was parsed into three datasets to allow for cross-validation of the measurement and structural models, following criteria set forth by Kroonenberg and Lewis [33]. The data was parsed using a random selection process, SAS 8.0. (SAS Institute, Cary, North Carolina). This allowed the measurement model to be tested (dataset #1 n=658), modified when necessary, and tested on an independent sample (dataset #2 n=664). It also allowed for the structural model to be tested on an independent sample (dataset #3 n=1322) (Figure 2). The measures derived from measurement model testing and modifications are described in the external exogenous and health belief variables sections below. The reported reliability results for the model's latent variables are based on tests of the measurement model (using dataset #3). Latent variables in structural equation modeling represent variables that are not directly observed but are inferred, through a mathematical model, from other observed variables.

Figure 2.

Consort Diagram of Survey Sample – Goes about here

External (Exogenous) Variables

The study investigated how the latent variables (Initial Physician-patient Regimen Discussion, Follow-up Physician Communication, and Family Hardiness) modeled with health beliefs (e.g., Benefit and Risk of Regimen, Understand Regimen, Money for Regimen, and Intention) and medication adherence.

Initial Regimen Discussion

The Initial Regimen Discussion scale was based on how the physician initially discussed the regimen with the patient (see Table 2 for variable description). The confirmatory factor analysis found that all variables were significant at p<.01 and the scale had a Cronbach's Alpha (α) = 0.71. Cronbach's Alphas (α) > 0.70 are considered desirable for structural equation models [34].

Table 2.

Mean, standardized factor loadings and error variances for external variables

	Meana	STDb	Standardized factor loadingc
Initial Regimen Discussion*
My doctor and I decided my regimen together	3.73	.626	.82
My doctor and I discussed side effects	3.72	.604	.98
My doctor and I have a follow-up plan	3.52	.803	.82
Follow-up Physician Communication
Comfortable talking with doctor	3.28	.666	.82
Doctor responds to my questions	3.02	.787	.80
I am able to reach my doctor	3.43	.641	.78
Family Hardiness
In long run good things are balanced by bad	3.43	.718	.67
We are strong in face of big problems	3.56	.546	.97
I trust things will work out in difficult times	3.57	.539	.95
We can count on each other	3.74	.470	.89
We do not feel we can handle another problemd	3.40	.837	.22
We believe things will work out	3.73	.485	.95
We strive to help each other	3.67	.498	.98
We work together to solve problems	3.60	.557	.94

^*questions paraphrased

^a= Four-point scale ranging from 1 (strongly disagree) to 4 (strongly agree)

^b= Standard Deviation

^c= All standardized factor loadings are significant at a = .05.

^d= Reverse coded

Follow-up Physician Communication Pattern

The Follow-up Physician Communication Pattern scale was based on the physician's on-going relationship with the patient (see Table 2 for variable description). The confirmatory factor analysis found all variables were significant at p<0.01 and had a Cronbach's Alpha (α) = 0.75. Patients were in a position to evaluate follow-up physician communication patterns since they had been on their regimen for more than six months.

Family Hardiness was measured using the Commitment Family Hardiness scale (FHI) developed by McCubbin et al. [35] (see Table 2 for variable description). The McCubbin FHI scale was chosen, as opposed to another family resilience scale, because FHI had been negatively associated with family caregiver traits of sense of caregiver burden [36] and caregiver depression [37]. The FHI index has been validated in previous research with an internal consistency of (α = 0.81) [38]. The confirmatory factor analysis for this research yielded a Cronbach's alpha of (α = 0.80).

2.5 Independent Health Belief Variables

The latent variable labeled Benefit and Risk was a two-item latent variable based on a scale validated by van den Putte [39]. The variable was based on the outcome expectancy construct, where it is projected that the patient considers the benefit of the health behavior along with the risk, or negative consequences, when deciding to perform a health behavior. The benefit was assessed with the single-item measure “I believe the regimen has worked and met its goal” and the risk was assessed by “Side effects were present and prevented a valued activity.” These two factors had significant (p<.05) factor loadings in the confirmatory model of .594 for benefit of regimen and .428 for side effects.

Two single-item observed variables were included because of their importance in the ability of the patient to perform the medication regimen [38]. Understand Regimen was assessed by the measure “I completely understand my regimen” and having sufficient Money for Regimen was assessed by “I do not have adequate money to pay for my medications.” This variable was intended to determine the patient's ability to bear the medication cost or sustain the expenditure over time. The Money for Regimen variable was designated as a control variable and projected to affect behavioral intent as well as medication adherence behavior.

2.6 Dependent Variables

Two outcome variables were utilized. Behavioral Intent was applied because it is projected to be a precursor to health behaviors and adherence to medication regimen [23]. Behavioral Intent was a single-item observed variable of “Do you intend to take your medication exactly as prescribed?” Medication adherence was measured by collecting longitudinal medication refill information for the subject's targeted statin or diabetes medication. Data was collected from April 1 to September 30, 2003 using United Healthcare claims data. The adherence data was reported in the form of a mean medication possession ratio (MPR). MPRs divide days supplied by total days for a given period and have been correlated with clinical outcomes across several studies [40-42].

2.7 Data Analysis

Using the dataset reserved for structural model testing (n=1322), the study analyzed the structural parameters and a structural equation model (SEM) for model fit. Polyserial/polychoric correlation matrices were used in the analysis because the data is ordinal and non-normal [43]. Lastly, a mediation analysis, using the Sobel test [44], is used to determine if the external variables (Initial Regimen Discussion Physician Communication, Follow-up Physician Communication, or Family Hardiness) had a significant effect on behavioral intent or medication.

The SEM analysis was conducted on LISREL version 8.8 [45] using weighted least squares due to the non-normal data [46,47]. Several indices of fit were used. Tabachnik and Fidell [48] suggest the ratio of chi-square (X²) to degrees of freedom should be < 2 for good model fit. Values of the root mean square error of approximation (RMSEA) of 0.06 or less were used and considered indicative of good fit [49]. The remaining indices suggested by Kelloway [50] included: goodness of fit (GFI), adjusted goodness of fit (AGFI), normed fit (NFI) and comparative fit (CFI). Values >0.95 were considered to be a good fit for these indices [49].

3. Results

3.1 Demographic data

There were differences in the study between participants and non-participants (Table 1). The participants were older (by 1.9 years; p<0.001) and included more females (by 2.3%; P<.001). The subjects on statins and oral diabetic agents had similar demographic features. There were also no statistical differences in demographic features between the parsed datasets.

Table 1.

Sociodemographic Characteristics of the Participants at Baseline

Subject Characteristics	Respondent Sample (n=1322)	Non-Respondent Sample (*** = p<0.001) (n=3758)
Age (years) (Standard Deviation)	54.2 (9.79)	52.2*** (9.97)
Gender (% Females)	43.7	40.5***
Race (% Nonwhite)	8.7	NA
Education (years) (Standard Deviation)	13.6 (1.9)	NA
Hyperlipidemia Regimen	733 (55%)	2029 (54%)
Diabetic Control Regimen	589 (45%)	1729 (45%)
Medication Adherence (Standard Deviation)	82.5% (18.51)	78.8%*** (21.75)

3.2 External variables: perceptions of communication and family

All standardized factor loadings of each construct were significantly different than zero, indicating that the construct for all external variables achieved convergent validity. The standardized factor loadings and error variances of these variables are shown in Table 2. The constructs of the external variables fit data well, indicated that they achieved construct validity, and were found to be reliable. The Structural Equation Model (SEM) measurement model satisfactorily met all fit index requirements (Chi-Square Ratio (X²/df) = .90; RMSEA 0.03; AGFI = 0.99; and CFI = 0.99).

All variables in the physician communication scales had a relationship with health beliefs at p< 0.05 in the predicted direction except for a) initial regimen communication and Money (β = -0.34) and b) follow-up regimen communication and behavioral beliefs β = 0.05) (Figure 3). Initial physician-patient regimen communication pattern had significant relationships at p< 0.01 with Behavior Beliefs (β = 0.36) and Understand (β = 0.43). Family Hardiness had no relationships at p< 0.10.

Figure 3.

Structural Equation Model – Goes About Here

3.3 Structural model of medication health beliefs and adherence

The Structural Equation Model (SEM) generated a model that satisfactorily met all fit index requirements (Chi-Square Ratio (X²/df) = 1.67; RMSEA 0.53; GFI = 0.983; AGFI = 0.977; NFI = 0.964; and CFI = 0.977).

3.4 Mediation Analysis

The latent variable Family Hardiness was excluded from the mediation analysis of Medication Use since no significant paths were found between Family Hardiness and belief variables. These paths should be significant for mediation to occur [51]. The Sobel test [44] analysis of the physician communication variables only found Initial Physician Communication had significant mediation effects on intention (Table 3). An analysis of indirect effects discovered the path that followed Initial Regimen Discussion, to Benefit and Risk, to Behavioral Intent, explained 68% of the total indirect effects of Physician Initial Regimen Discussion on Behavioral Intent. The Sobel tests [44] between the physician communication variables and medication use found many significant effects. Yet, an analysis of indirect effects found the path that followed Follow-up Physician Communication, to Money, to Medication Use, explained 36% of the total indirect effects on medication use and 54% of the positive indirect effects on medication use (when the path between Initial Regimen Physician communication and money is excluded).

Table 3.

Matrix of Effects

Direct Effects	Unstandardized Estimate	Unstandardized Standard Error	Standardized Estimate	P-value
Initial Regimen Communication – Behavioral Beliefs	.23	.06	.38	.000
Initial Regimen Communication – Understand	.44	.11	.36	.000
Initial Regimen Communication – Money	−.44	.14	−.36	.001
Follow-up Communication – Behavioral Beliefs	.03	.03	.05	.562
Follow-up Communication – Understand	.22	.11	.18	.499
Follow-up Communication – Money	.55	.14	.45	.000
Family Hardiness – Behavioral Beliefs	.09	.03	.07	.078
Family Hardiness – Understand	−.02	.05	−.02	.638
Family Hardiness – Money	.10	.06	.07	.092
Behavioral Beliefs – Intention	.88	.11	.43	.000
Understand – Intention	.09	.03	.09	.006
Money – Intention	.12	.05	.12	.017
Money – Medication Use	.25	.05	.25	.000

Indirect Effects	Test Statistic	Std. Error	p-value	% Mediation Effect
Initial Regimen Communication – Behavioral Beliefs – Intention	3.43	.059	.000	68%
Initial Regimen Communication – Understand – Intention	2.40	.017	.017	14%
Initial Regimen Communication – Money – Intention	1.23	.014	.220	8%
Follow-up Communication – Behavioral Beliefs – Intention	.990	.027	.321	NS*
Follow-up Communication – Understand – Intention	1.66	.012	.096	NS
Follow-up Communication – Money – Intention	1.95	.032	.061	NS
Initial Regimen Communication – Behavioral Beliefs – Intention – Use	9.85	.104	.000	16%
Initial Regimen Communication – Understand – Intention – Use	8.16	.089	.000	3%
Initial Regimen Communication – Money – Intention – Use	9.93	.030	.000	4%
Initial Regimen Communication – Money – Use	2.99	.044	.003	29%
Follow-up Communication – Behavioral Beliefs – Intention – Use	9.64	.031	.000	2%
Follow-up Communication – Understand – Intention – Use	9.96	.050	.000	5%
Follow-up Communication – Money – Intention – Use	9.88	.062	.000	5%
Follow-up Communication – Money – Use	9.43	.045	.000	36%

^*NS = No significant mediation effect

4.1 Discussion

Annually, non-adherence to evidence-based medication regimens alone generates over $90 billion in healthcare costs in the United States [52]. This is just one example of the poor maintenance of a desirable behavior. More than 40% of adopted evidence-based preventive health practices fail to achieve behavioral maintenance once adopted [10,53]. http://healthaffairs.org/blog/2012/08/28/seizing-the-opportunity-to-improve-medication-adherence/ Despite the need, the scientific community is unable to define a valid model of sustained use behavior [54]. Furthermore, healthcare providers are not sure how to prevent behavioral relapse to pre-existing behaviors. For unhealthy behaviors such as poor medication adherence, it has been surmised that relapse is due to social and environmental forces beyond the control of the clinical professional [55]. Based on this presumption, the likelihood of regimen adherence is considered a matter of chance. However, this research suggests that assumption may be inaccurate for patients on hyperlipidemia or diabetes medication regimens. The research demonstrated that factors related to physician communication patterns provide the foundation for a model of long-term adherence to a medication regimen. The model had strong fit indices (GFI = 0.982, AGFI = 0.976; NFI = >0.960; and CFI = 0.974) and good fit with a root mean square error of approximation (RMSEA) score of 0.053. These results suggest both individual cognitive beliefs and physician communications play an active role in a patient's maintenance to a medication regimen.

4.2 Practice Implications

Family hardiness contributes to a model of sustained use of a regimen, but does not contribute directly or indirectly to intention or actual behavior. Family hardiness and resilience are posited to assist families encountering adversity or a crisis [56]. Families with a family member undergoing a long-term medication regimen may not be under enough adversity to warrant the use of the family hardiness construct to explain medication adherence behaviors.

Two general conclusions emerged from the investigation of physician patient communication and adherence to a medication regimen. First, a physician's initial regimen communication patterns have an impact on a patient's medication beliefs and intent. These initial communication patterns work to build the behavioral belief that the benefit of the medication regimen will outweigh negative consequences related to side effects. This in turn gets the patient “ready” for the change by bolstering intent to adhere to the regimen. This reported research builds upon the research of Bultman and Svarstad [57] that found the physician's initial communication style positively influences initial beliefs about the medication. Their study, as with the reported research, assessed health beliefs related to perceived benefits of the regimen as well as perceived side effects. Moreover, our research demonstrated how the initial regimen discussion affects the patient's on-going regimen beliefs for a regimen for greater than six months. The applied use of these findings suggests physicians should stress the potential benefits of the regimen, how to avoid or address common side effects, and the importance of adhering to the regimen during the initial regimen communication.

The second general conclusion is that the Follow-up Physician Communication and Money for Regimen variables affect actual medication use behavior. This occurs through an indirect effect mediated by the variable money (or perceived ability to pay for a regimen). The combination of follow-up physician communication and the money variable did not impact the patient's initial belief they will take a medication, but has a direct relationship on extended medication use. This finding underscores the importance of follow-up communications to “get the patient through the change.” Furthermore, once a person has intent to make a change, barriers may arise, and it is the follow-up physician communication that can assist patients in successfully overcoming these barriers. Trends in physician communication describe how concordance (or alignment) between the patient's beliefs and needs and the regimen the physician prescribes can enhance regimen adherence [58]. The follow-up physician communication with the patient should determine if the medication regimen ought to be adjusted because of patient concerns about ability to pay for their medication.

Motivational Interviewing is an evidence-based communication practice for discussing issues during the initial regimen discussion or follow-up appointments [59]. This approach can be used to develop patient-derived motivation for the regimen and strategies to overcome money related and other barriers to regimen adherence. For the physician, the additional conversation with the patient could be brief and address a common concern of health insurers: poor medication adherence. The benefit of these communication behaviors could be a 10% improvement in patient adherence, since patients that rated physician communication high in the study had a 10% higher adherence rate than those that did not.

4.3 Methodological Considerations

Several aspects of the research design strengthened the model's validity. First, medication use was considered an outcome variable in the structured equation model (SEM) rather than just behavioral intent. Many research trials of behavior change use behavioral intention as the only outcome variable because intention data is easier to secure [60]. Theoretically, intention is considered a necessary, but not sufficient, predictor of use, since many well-intentioned individuals cannot sustain a change in behavior [61]. Second, this research design used medication refill records as the measure of adherence, rather self-reported adherence data. Cognitive behavioral and medication adherence research tends to rely on self-reports for outcome variables, despite evidence to suggest the vulnerability of such data due to social desirability biases [62]. The positive bias of self-reported outcome variables has led to a “self-report effect” in cognitive behavioral research that results in over-reporting of the explained variance between predictor and outcome variables by as much as 50% [61]. Third, the model was tested on a larger than normal sample (> 1300). Medication adherence trials often have samples of <200 [63]. Hence, the statistical power is often not of the same magnitude as the sample in this trial of >600. Large samples also afford the use of structural equation models (SEMs).

There are some limitations to this study. First, this research was limited in the constructs used to assess physician behavior and family hardiness. 1) Other measures of social and family support should be considered in future research. For instance, research using the ENRICHD Social Support Instrument (ESSI) has demonstrated that having someone to talk to, assist with daily chores, and provide emotional support all increase medication adherence for those with high health literacy [64]. Another trial demonstrated that social support could be most effective if it focuses on a patient's self-care skills [65]. 2) Family composition was not studied in this analysis. The past pattern for family hardiness and resilience research has been to have the patient self-determine their definition of “family” as they answer the survey questions [66,67]. The study did include members of family plans: meaning a spouse or children were also insured with the study subject. 3) In addition, the physician communication patterns are based on self-report. Hence, these measures are limited to patient perceptions of physician communication and are subject to biases inherent to patient personality, liking of physician, etc. This study gives indications of how patients perceived this relationship, and a study that incorporated audiotapes of physician patient communication could be used to compare patient perceptions to a more objective measure. 4) There are other scales and constructs of physician communication based on patient-perceived empathy [68] and provider participatory decision-making [69]. This study chose to follow only the constructs prioritized by the expert panel. Future research can expand upon these findings by including additional physician communication constructs. 5) Lastly, the medication adherence rates in the study are higher than the 40–60% typically found in medication adherence studies [7,8], but are common for medication adherence rates generated from health plan pharmacy claims data [70,71]. It should noted that refill data is a measure of maximum potential adherence, due to the measure being based on refills filled versus actual medication taken, and refill data rates have been linked to health outcomes [41].

Another limitation was that the study had a nearly 60% non-inclusion rate once all factors for exclusion were considered. Non-response is a common, recognized limitation of survey methodology; our response rate of 42.2% is within expected range [72]. The non-response bias due to fewer responses from younger and male patients is found in other studies [72,73]. The research also does not incorporate disease severity. The impact of disease severity has been mixed, with patients with less severe diseases being more adherent to their medications and those with more severe diseases being less adherent [74]. A greater understanding of the relationship between severity of disease and medication regimen beliefs is also warranted. Lastly, this research only addressed two disease conditions and is generalizable only to hyperlipidemia and diabetes.

Conclusion

This research only scratches the surface of the influence of social variables on maintenance of one long-term health behavior: medication adherence for a chronic disease prevention regimen. It supports the important role the physician plays in health behavior maintenance. Future research should study the effect physicians have on other health behaviors. Moreover, this model provides a framework for studying behavior maintenance that may be generalizable to other contexts.