The relationship between tendency to adhere to provider-recommended treatments and subsequent pain severity among cancer patients has not been previously examined. This is an important oversight, given the generally suboptimal state of cancer pain control,1
the high societal prevalence of serious misgivings about the safety of analgesics commonly prescribed to treat cancer pain,1
and the apparently low levels of adherence to recommended cancer pain treatment regimens.3
Our findings begin to address this important yet previously neglected relationship.
In a secondary observational analysis of prospective data from patients enrolled in an RCT of a cancer pain coaching intervention, we found patients’ general tendency to adhere to provider-recommended treatments at baseline predicted cancer pain severity over 12 weeks’ follow-up in a model adjusting for basic sociodemographic characteristics, RCT-related variables, and baseline pain (Model 1). The sociodemographic characteristics in this model (age, gender, race/ethnicity, and education) were included because each has been associated with pain15
in prior studies. Particularly in the context of the observed significant association between general tendency to adhere and pain severity, we considered that including baseline pain in the model, although conservative, might constitute overadjustment, as some of the influence of adherence tendency on pain severity would likely have been exerted at the start of the study. Consistent with this reasoning, in an analysis omitting baseline pain from Model 1, there was an additional 0.1-point decrement in adjusted mean pain severity associated with a one standard deviation increase in tendency to adhere. On the other hand, not adjusting for baseline pain may represent underadjustment. Thus, the “true” point estimate of the association between baseline general treatment adherence tendency and subsequent pain severity may lie somewhere between these values.
These findings are broadly consistent with those of a large number of prior studies employing a variety of adherence measures, including measures of general tendency to adhere, and involving patients with diseases other than cancer and outcomes other than pain severity.8
Additionally, a cross-lagged model analysis revealed no significant association between baseline pain severity and 6 week adherence tendency, contrasting with the aforementioned significant association between baseline adherence tendency and 6 week pain severity, and baseline tendency to adhere parameter estimate was significantly larger than the 6 week tendency to adhere parameter estimate. These findings suggest that a simple “halo effect” due to underlying patient dispositional characteristics influencing both adherence tendency and pain severity is unlikely to account for the observed association between adherence tendency and subsequent pain severity. Rather, the temporal sequencing of the relationship (baseline adherence tendency associated with subsequent pain, but baseline pain not associated with subsequent adherence tendency) is consistent with a causal relationship. Thus, measurement of baseline tendency to adhere to recommended treatments may be useful in predicting pain severity over time among cancer patients. Furthermore, studies involving individuals with noncancer health conditions suggest “tendency to adhere” is a more mutable characteristic than the label might imply, and so might represent a useful therapeutic target. Future RCTs of interventions expressly designed to improve cancer pain control through salutary effects on tendency to adhere would be helpful to more definitively investigate these possibilities.
To further explore the relationship between adherence tendency and pain severity, we also created an expanded model that included all Model 1 variables as well as several personal characteristics likely to influence the tendency to adhere, pain severity, or both: partner status, mental and physical health status, and pain control self-efficacy (Model 2). These additional variables were included in the model based on previously demonstrated associations with pain14
as well as with adherence.23
In this model, patient tendency to adhere was no longer significantly associated with follow-up pain severity, and the inclusion of two variables in particular (physical and mental health status) appeared to account for this finding. This is again generally consistent with prior research involving patients with health conditions other than cancer, in which lower mental health status28
and, particularly for more burdensome health conditions, lower physical health status34
have been associated with lower adherence estimates, including lower tendency to adhere as assessed with the MOS general adherence measure.33
One potential interpretation of this finding is that the general tendency to adhere to provider-recommended treatments influences subsequent pain severity among cancer patients, and that physical and mental health statuses mediate this effect. Individuals who assess themselves as being more likely to adhere to provider-recommended treatments may actually be more adherent to specific pain treatments (not measured in the current study). Better adherence could improve physical and mental health status, in turn leading to less “actual” pain (due to reduced physical illness burden) and reduced tendency to report severe pain (due to improved mental health and better coping). Alternatively, physical and mental health status may confound the relationship between tendency to adhere and subsequent cancer pain severity. Such confounding could reflect the influence of relatively fixed underlying dispositional characteristics (eg, optimism on the “positive” side and neuroticism on the “negative” side). Such characteristics are likely to influence patient responses to each of the relevant self-report measures (adherence tendency, health status, pain severity), which seek to capture respondent standing on conceptually different yet, in reality, partially overlapping constructs. Including the SF-12 in our models might also result in overadjustment, as the measure includes one item concerning pain (asking respondents to rate how much pain interfered with their normal work during the preceding 4 weeks). Another potential source of confounding is that individuals with poor health status, reflecting poor control of their health conditions, may also have more pain, which might impair their ability to adhere to recommended treatments. In the context of observational data such as those available here, the distinction between mediation and confounding cannot be determined,48
yet the research and clinical implications of each explanation would be somewhat different. Future studies will be required to clarify this issue.
Our study had a number of strengths, including a prospective design, the use of rigorous analytic methods, and a sample that included patients with a range of cancer types and severities. Our study also had some limitations. Though our analyses examined temporal associations between adherence and pain severity, causality cannot be inferred from these observational relationships. The study data were also drawn from a sample of English-speaking cancer patients cared for by cancer specialists in one metropolitan area, so the findings may or may not be generalize to other patients, physicians, or geographic regions. Finally, the parent RCT from which the study data were drawn included a single adherence measure assessing patients’ global tendency to adhere to provider-recommended treatments. Thus, we were unable to examine the relationship between actual patient adherence to specific recommended pain treatments, such as individual prescription or over-the-counter pain medications, and pain severity. The findings of analyses similar to ours but that focus on the effects of adherence to specific recommended treatments might differ.
In conclusion, in a secondary analysis of data from an RCT of cancer patient coaching intervention, we found that baseline general patient tendency to adhere to provider-recommended treatments was significantly associated with cancer pain severity over 12 weeks when adjusting for basic sociodemographic characteristics, RCT-related variables, and baseline pain. Furthermore, the results of a cross-lagged analysis suggested this association is unlikely to merely reflect underlying patient dispositional characteristics that influence both adherence tendency and pain severity. If confirmed in other studies designed specifically to examine causality, this finding would suggest that general tendency to adhere to provider-recommended treatments might be useful as a predictor of subsequent cancer pain severity and as a target of interventions to improve cancer pain control. Tempering this finding somewhat, we also found that after further adjusting for health status, the association between adherence tendency and pain severity was no longer significant. Thus, health status either mediated or confounded the adherence tendency/pain severity relationship, with each of these potential interpretations having somewhat different research and clinical implications. Further prospective studies designed to clarify the nature of the relationships among tendency to adhere, health status, and pain severity among cancer patients appear warranted.