Nonadherence to Medication Four Years After a First Episode of Psychosis and Associated Risk Factors

Nonadherence to antipsychotic medication among people with schizophrenia is associated with relapse rates of almost 80% after one year without medication and 96% after two years. Nonadherence results in worse illness outcomes, more readmissions, and increased costs to society ( 1 ).

Putative correlates of nonadherence include age, gender, symptom severity, substance misuse, and attitudes toward illness and treatment ( 2 ). Identifying factors at first presentation that predict nonadherence in the future increases our knowledge of risk factors and has practical clinical implications. There are few prospective follow-up studies of patients with first-episode psychosis from defined geographical regions ( 3 , 4 ), and there are even fewer studies reporting the relationship between the duration of untreated psychosis and medication adherence ( 5 , 6 ), despite the recent focus on the assessment of potential confounders of the relationship between duration of untreated psychosis and outcome.

In this prospective cohort study we measured the rate of adherence among patients four years after a first episode of psychosis. We tested the hypothesis that factors at inception that have been linked to nonadherence in other studies would predict nonadherence in this study in a group without previous exposure to antipsychotic medication. Another aim was to explore the relationships between concurrent patient outcomes at four years and nonadherence.

Methods

Participants were drawn from a prospective cohort study of first-episode psychosis based in an urban region in Ireland with a population of approximately 172,000. The study received approval from the local ethics committee. We included all patients presenting with a first episode of psychosis who had not been prescribed antipsychotic treatment for more than 30 days between February 1995 and February 1999. Written informed consent was obtained.

At inception, we performed a Structured Clinical Interview for DSM-IV to determine diagnosis. We evaluated patients with the Global Assessment of Functioning (GAF) and derived data about alcohol and drug misuse.

We assessed symptomatology using the Positive and Negative Syndrome Scale (PANSS). The Drug Attitude Inventory (DAI) provided information about patients' attitudes toward treatment. We used the Birchwood self-report questionnaire and item g12 on the PANSS to measure insight.

We used the Beiser scale to measure the duration of untreated psychosis. The individual and his or her family were interviewed and a duration in months was obtained from each. We came to a consensus figure when there was a discrepancy between the individual and the family reports based on the sources available to us. We defined duration of untreated psychosis as the period between the onset of the first psychotic symptom and commencement of any antipsychotic treatment.

We used the Premorbid Adjustment Scale to assess premorbid social adjustment during two stages of early life: age five to 11 (PAS1) and age 12–16 (PAS2). Because the PAS2 interview is often conducted at the age of prodrome of psychosis, we used only the PAS1 scores in our analyses.

We repeated all of these assessments at four years, with the exception of the Beiser and the Premorbid Adjustment Scale. Each interviewer was blinded to all previous assessments. The methods and measures used have been described in detail elsewhere ( 7 ).

We measured adherence to medication among patients prescribed antipsychotic medication at four years using the Compliance Interview—a validated six-item interview examining patients' adherence over the past month ( 8 ). Patients are rated on a 4-point scale; 1 indicates compliance of 0%–24%; 2, 25%–49%; 3, 50%–74%; and 4, 75%–100%. We defined adherence as a score of 4 and nonadherence as a score of 3 or less, an approach consistent with other studies ( 9 ).

We interviewed 171 people at baseline. Patients with a diagnosis of psychosis owing to a general medical condition were excluded, leaving 166 participants. We interviewed 132 patients at four years. Twenty-six of these were no longer being prescribed antipsychotic medication, and we excluded them from our analysis. Two of the 106 remaining had missing baseline assessments and were excluded from our multivariate analyses. At presentation 20 (19%) of the group with adherence measures at four years refused consent for the family interview, which meant we did not have premorbid social adjustment scores for these. We therefore carried out a primary analysis of the 84 patients to control for premorbid adjustment in the analysis of the relationship between duration of untreated psychosis and adherence. We performed a secondary analysis in which we did not control for premorbid adjustment for 104 participants.

There were no significant differences between the group for which we had complete four-year adherence data (N=104) and the incomplete data group in terms of gender, age, duration of untreated psychosis, premorbid social adjustment, years of education, psychopathology, substance misuse, or global functioning at presentation.

We used the SPSS, version 15.0, for statistical analyses. We log-transformed the duration of untreated psychosis because it was positively skewed. We used chi square tests, Student's t tests, and Mann-Whitney U tests to establish that there were no differences at first presentation with respect to all relevant sociodemographic and illness parameters between patients with adherence data at four years and patients without. We carried out similar univariate analyses to compare baseline and four-year sociodemographic and clinical variables between the adherent and nonadherent groups at four years.

To identify independent predictors of nonadherence four years later, we used binary logistic regression modeling with the following independent predictors: age, gender, duration of untreated psychosis, premorbid social adjustment, treatment setting, alcohol or drug misuse at time of presentation, positive symptoms, negative symptoms, insight, and global functioning. We repeated the analysis with the removal of premorbid social adjustment.

Results

Eighty-one patients (76%) were adherent to medication and 25 (24%) were not ( Table 1 ). Univariate analysis showed that participants who were nonadherent at four years were significantly more likely to have been misusing substances at initial presentation (p=.02) and concurrently (p<.01). In addition, they were significantly more likely to have been readmitted to a hospital since their first episode of illness (p=.01). They also had a significantly longer duration of untreated psychosis (p=.04). The median duration of untreated psychosis for those adherent at four years was four months, and it was 18 months for those who were nonadherent. This association remained significant when we excluded those with a diagnosis of an affective psychosis (p=.03).

Enlarge table

There were significant associations between nonadherence and lower Birchwood scores (p=.02), lower GAF scores (p<.01), and lower DAI scores (p<.01) at four years. PANSS scores were significantly higher in this group, including g12 scores, in particular on lack of judgment and insight (p<.01).

Logistic regression analysis revealed that alcohol or drug misuse at presentation (adjusted odds ratio [AOR]=6.9, 95% confidence interval [CI]=2.0–23.7, p=.02) and longer duration of untreated psychosis (AOR=2.5, CI=1.1–5.9, p=.04) predicted nonadherence at four years. The model was significant ( χ² =15.5, df=6, p=.02) and accounted for 25% of the variance (Nagelkerke R ² =.25). Its percentage accuracy in classification was 73.8%. These results remained significant when PAS1 was removed from the model: alcohol or drug misuse (AOR=3.7, CI=1.3–10.8, p=.01) and longer duration of untreated psychosis (AOR=2.1, CI=1.1–4.2, p=.04).

Further analysis showed that duration of untreated psychosis was associated with symptom severity at four years ( β =.35, p<.01) but not with hospital readmissions ( β =.01, p=.85). The addition of adherence to the linear regression model did not significantly reduce this relationship. Both adherence ( β =.27, p<.01) and duration of untreated psychosis ( β =.31, p<.01) were significantly associated with outcome.

Discussion

Studying patients from their first episode of psychosis before neuroleptic treatment and measuring their adherence to medication at a later stage afforded us an opportunity to identify risk factors from time of first presentation for future nonadherence. Our results replicate findings from other studies in terms of cross-sectional associations between more symptoms, less insight, and negative attitudes toward treatment and nonadherence. However, we did not find that these variables at first presentation of psychosis predicted nonadherence four years later. Perhaps this is a result of a type II error or a reflection of an individual's changing adherence status over time. Our finding that alcohol or substance misuse at inception predicted nonadherence four years later is unsurprising and similar to findings of our previous investigation of adherence at six months ( 2 , 3 , 9 ).

The association between longer duration of untreated psychosis and later nonadherence to antipsychotic medication at four years has important implications. Adherence and duration of untreated psychosis have been shown to be important predictors of outcome in psychosis ( 2 , 10 ), which poses the question of whether or not the relationship between duration of untreated psychosis and outcome may, at least partly, be mediated by medication adherence. This was not the case in this study, and both were associated with symptom severity at four years.

This finding implies that the relationship between duration of untreated psychosis and adherence is distinct from its relationship with symptomatic outcome. One possible interpretation is that common factors result in both prolonged duration of untreated psychosis and reluctance to adhere to treatment, such as reluctance to seek help, denial, embarrassment, and distrust of medical treatment ( 11 ). Alternatively, a gradual development of illness may allow time to adapt to symptoms, or perhaps treatment delay causes nonadherence.

Studies reporting the relationship between duration of untreated psychosis and adherence are limited and results are conflicting ( 5 , 6 , 11 , 12 , 13 ). To our knowledge a long-term association between duration of untreated psychosis and adherence has not been demonstrated in a representative first-episode sample.

The main strengths of this study are its sampling and length of follow-up. All inpatients and outpatients with a first episode of psychosis who were referred to a geographically defined area in Ireland were identified. There were no exclusion criteria relating to age or substance misuse and there were no initial refusals to participate, making the sample representative and generalizable. We used standardized assessment tools and had a consistent measure of duration of untreated psychosis. There were no differences at inception between those measured for adherence at four years and those not, which limited potential bias.

A major limitation was our measure of adherence. Recent research has established the problems with self-report measures of adherence ( 14 ). They are susceptible to error because they can be easily distorted by patients, who have a tendency to overreport, and the low rate of nonadherence in this study could be due to this. Unfortunately, these findings had not been published before we conducted our follow-up assessment; nonetheless, an additional direct or objective measure, such as pill count, urine or plasma analyses, and electronic monitoring and pharmacy refill records, would have added to the validity of our results. However, these tests are not without their own limitations. Self-report measures are simple, inexpensive, and therefore useful in a clinical setting, and a recent study indicated close agreement between electronically monitored ratings of adherence and adherence estimated by research assistants ( 15 ). We had a naturalistic cohort of participants who were followed independently from their clinical teams, which may have decreased the likelihood of underreporting somewhat.

Another important limitation is the fact that adherence is measured at a single cross-sectional time point, which gives us only a snapshot of an individual's adherence. Repeated measures of adherence status at intervals over the four years would allow identification of a subgroup with persistent nonadherence patterns; however, this was not possible because this was a follow-up study at a single time point.

Finally, as is commonly encountered in cohort studies, a considerable number of patients were lost to follow-up, and this may have led to an overestimation of adherence, with the supposition that those harder to follow are less likely to adhere to medication. However, attrition was less than 30%. Also, because of missing data, our sample size may have been underpowered to detect some associations. We may have falsely discounted the predictive validity of DAI, Birchwood, and PANSS positive symptoms at first presentation.

Conclusions

Despite these limitations, our findings support previous cross-sectional research showing that among individuals with a psychotic illness identifiable groups are likely to be nonadherent to their antipsychotic medication. Furthermore, this study suggests that at first presentation of psychosis individuals with coexisting substance misuse and longer duration of untreated psychosis were less likely to adhere to medication in the future and may represent a group who require interventions targeted toward improving adherence. Further research into the relationship between duration of untreated psychosis and adherence is warranted, with more detailed exploration of individual factors that contribute to longer duration of untreated psychosis. Such research may establish whether there are common factors between those who take longer to present with psychosis and patients who do not take their medication and whether early intervention may have a role to play in reduction of nonadherence rates.

Acknowledgments and disclosures

This study was supported by grant RP-2007-190 from the Health Research Board of Ireland and the Stanley Medical Research Institute.

The authors report no competing interests.