Abstract
Over the last five years the prognosis for patients with HIV infection has improved dramatically with the introduction of several classes of powerful antiretroviral (ARV) medications into routine clinical use [1]. By using combination triple-drug treatments, known as highly active antiretroviral therapy (HAART), it is now possible to attain durable viral suppression in most antiretroviral drug-naïve patients [2]. This permits reconstitution of the immune system, with gradual increase in CD4 cells and other immune-competent cells, and improved physical health.
However, due to its rapid mutation, the virus is able to develop resistance to antiretroviral drugs. The consequences of viral drug resistance include inability to suppress the virus, disease progression and possible transmission of drug-resistant viral strains to others [3].
Close adherence (or compliance) to the prescribed antiretroviral regimens is essential, if viral replication is to remain suppressed and viral drug resistance prevented [2–4]. A ‘drug holiday’ of even several days frequently leads to a substantial increase in plasma viral RNA level [5]. Viral replication allows mutation, so viral resistance may rapidly emerge.
Most recent studies have suggested that patients need to take at least 90% of their antiretroviral medication to completely suppress viral replication [6, 7]. Adherence to HAART is difficult. Patients must take several medications, in large numbers, several times a day, either before food, with food or after food. The drugs have a variety of unpleasant, often debilitating, and at times potentially life-threatening side-effects, which often result in medication cessation.
Medication adherence has been studied extensively since the mid-1970s in general medical patients. For example, the minimum level of adherence required to achieve treatment goals ranges from 30% of prescribed penicillin for rheumatic fever prophylaxis to nearly 100% for oral contraceptives [8, 9]. Factors associated with decreased adherence include disease chronicity, complex drug regimens and lack of warmth in the patient-doctor relationship [10, 11]. Adherence is also influenced by the patient's beliefs about illness [12, 13]. Depression has been found to correlate with poor medication adherence in patients with asthma [14], diabetes mellitus [15] and following kidney transplantation [16].
There is no ideal method of assessing medication adherence [17]. Methods used include self-report and interview, clinical response, biochemical markers such as drug levels, pill count, and computerized compliance monitoring using the Medication Event Monitoring System (MEMS). While MEMS is considered the most reliable method [18], it is expensive and intrusive. It requires the patient to use only specially prepared medication containers, which precludes the use of adherence aids such as the dosette box or the Webster Pack, in which all doses for the week are placed in push-out foil sealed cells in a single-use dosette box prepared for the week by a pharmacist. Such adherence aids greatly help patients who are forgetful or cognitively impaired, and facilitate supervised dosing. Hence, use of the MEMS may make it impossible for the most vulnerable patients to comply with the medication regimen. Because of cost and convenience, many studies of medication adherence in patients with HIV have used principally self-report or interview to provide estimates of adherence [3, 6, 19].
Characteristics of patients with HIV that have been shown to impact significantly on adherence include depression [6], African-American ethnicity [20], poor housing [21], past intravenous drug use [19] and negative beliefs about the medication [19]. Several studies have been of cross-sectional design [6, 19, 22], others longitudinal in varying degrees [3, 7, 20, 21, 23, 24].
There is no published Australian study focusing on psychological distress and adherence of patients with HIV to antiretroviral medication. The aims of the study were: to determine if there is a relationship between psychiatric morbidity and poor antiretroviral medication adherence in a hospital clinic population; to audit selfrated adherence to medication; and to test the value of the General Health Questionnaire (GHQ) in predicting self-reported adherence. The chief hypothesis tested was that psychological distress, as measured by the GHQ, predicts poor adherence to medication.
From past unpublished surveys of the clinic where this study was conducted, it is known that the principal risk factor for HIV infection is male-to-male sex, while intravenous drug use is a risk factor in a significant minority.
Methods
Participants
The Immune Monitoring Clinic is an outpatient clinic for medical management of HIV-positive patients in a university-affiliated innercity public hospital. All English-speaking patients attending the Clinic on the second or subsequent visit, were eligible for the study. After explaining the study, the clinic nurse invited patients who were due to see a doctor that day, to participate. If the patient agreed, he/she was given a numbered survey form with a written explanation of the study. Enrolment for the study commenced at the start of November 2000 and ended in mid-February 2001.
The patient completed the survey in private, and the patient's HIV physician completed a matched one-page survey separately. Approval was obtained from the hospital Ethics Committee before the study began.
The survey
Data on age, sex and number of years of living with HIV were obtained. Other information included the patient's current drug and alcohol use, health beliefs such as perceived severity of current and future illness, current social support and attitude to medication on a four-point scale, and whether he/she had been admitted to a psychiatric ward or taken psychotropic medication in the past. Disease knowledge was assessed by asking the patient to recall their last CD4 cell count and viral load. All of these responses were dichotomised.
Adherence was assessed using a combination of two methods of self-report. The first involved asking the patient the number of tablets missed the previous day and during the previous week. If the patient admitted to missing more than 10% for either period, adherence was considered poor, in line with the studies of antiretroviral effectiveness using 90% adherence as the benchmark. The second method used a modified Morisky medication adherence questionnaire [25], as shown below. If the patient answered ‘yes’ to Questions 2, 3 or 4 then adherence was considered poor. An affirmative to Question 1 alone was not considered indicative of poor adherence.
Modified Morisky medication adherence questionnaire
The original version used ‘careless’ in Question 2 – we adapted this to avoid the possible pejorative meaning of ‘careless’. The original questionnaire differentiated between ‘high’, ‘medium’ and ‘low’ compliance, based on total ‘yes’ scores of 0, 1 or 2, and 3 or 4, respectively. The modified version described has not been compared with a ‘gold standard’ such as the MEMS.
The above two adherence measures were combined: poor adherence on either measure was deemed to show poor adherence overall. Good adherence was required on both measures to show good overall adherence.
Psychological distress was assessed using the 28-question General Health Questionnaire (GHQ-28) [26]. The GHQ has been used extensively in general practice research and in specialist medical patients [27, 28]. It is designed to detect significant anxiety, depressive and undifferentiated psychiatric syndromes [23]. The GHQ was scored 0-0-1-1, and a cutting score of 4/5, as recommended for general practice patients, was used to determine ‘case’ [26].
Somatic symptoms on the GHQ may be present due to physical illness, rather than psychological morbidity, hence the cutting score has not been defined for patients suffering various chronic medical disorders. Goldberg has recommended pilot validation studies to establish an appropriate score, in order to balance sensitivity and specificity [28]. Ideally, a pilot validation study comparing GHQ with results of diagnostic interviews may have provided guidance to the most appropriate cutting score.
Statistical processing
The Statistical Package for the Social Sciences (SPSS), Version 10, was used for statistical analysis. Pearson's chi-squared test (χ2) was used in the univariate analysis for categorical data where appropriate, and Fisher's exact test was applied for small numbers. One-way analysis of variance (ANOVA) was employed for multiple normally distributed groups of numerical data. All p-values were computed using two-tailed tests of significance.
Logistic regression was used for clinically important variables found in univariate analysis to be associated with poor adherence at p ≤ 0.20. Adjusted odds ratios and 95% confidence intervals were calculated by this method. This approach to logistic regression is described by Samet [19]. Variables that were highly intercorrelated, or which involved low numbers of patients and hence not suitable for χ2 analysis, were excluded from the logistic regression.
Results
Ninety-three clinic patients were invited to participate in the survey. Five declined, three patients did not return the survey, three did not complete the General Health Questionnaire and three failed to complete the adherence questions. This left 79 participants.
Patient characteristics
Of the 79 subjects, 75 (95%) were male, one was transgender and three were female. Mean age was 42.5 years (range 24–67), and mean duration of known HIV seropositivity on self-report was 8.6 years, range 0.4–20. Nineteen subjects (26%) had CD4 cell counts below 200 cells/μL, 34 (46%) between 200 and 499 and 21 (28%) above 500. Almost half (42%) had no detectable virus at last measurement. Thirtyone (39%) described illicit substance use, and seven (9%) intravenous use in the previous month. Six (8%) reported hazardous alcohol intake (females over 140 g per week, males over 280 g per week).
Sixteen subjects (24%) recalled neither their latest CD4 count nor their viral load correctly – their disease knowledge was considered poor. Four (5%) perceived their current illness as severe, 15 (21%) believed that they would be severely ill in the future, 11 (14%) expressed scepticism about the ability of medication to keep them well, and 67 (88%) were satisfied with their level of social support. Thirtytwo subjects (41%) had taken ‘medication for nerves’, and five (6%) reported a past psychiatric admission. 35 (44%) were identified as ‘cases’ using the GHQ-28 with the cut off score at 4/5.
Sixty-six patients (84%) were prescribed antiretroviral treatment, but only 36 (46%) were prescribed prophylactic anti-infective medication. Antiretroviral medication adherence was determined as good in 68% (45/66) of those prescribed antiretrovirals. Physicians completed and returned 66 (84%) one-page forms. Fifty-eight of these patients were taking antiretroviral medication, of whom 35 (60%) were considered by their doctors to have good medication adherence.
Variables associated with poor adherence: univariate analysis
As shown in Table 1, poor antiretroviral medication adherence was found to be significantly associated with past psychotropic medication, psychiatric morbidity as detected by the GHQ, and scepticism about effectiveness of antiretrovirals.
The following factors were not significantly associated with poor medication adherence: substance use (hazardous drinking, illicit substance use and recent intravenous drug use), disease knowledge (knowing one's most recent CD4 cell count or viral load), perceived severity of current and future illness, perceived quality of social support, age, duration of known HIV seropositivity, current CD4 cell count, current viral load, past psychiatric admission and consumption of prophylactic anti-infective medication.
The physicians’ predictions of adherence revealed an overall prediction accuracy of 77%, when compared with the patients’ self-report. Kappa was 0.48, indicating moderate agreement. The doctors’ overall sensitivity in predicting poor adherence was 0.69 and specificity 0.81, using the patients’ self-report as a ‘gold standard’.
Variables predicting poor adherence
Table 2 shows the effect of varying the cutting score for the GHQ-28 from 4/5, as was used in the analysis, up to 11/12. Despite the reduction in the number of patients identified as ‘cases’, the association with poor adherence remains statistically significant in this posthoc analysis.
GHQ cutting score and poor adherence to antiretroviral medication
Variables associated with not being prescribed medication: univariate analysis
Thirteen patients were not prescribed antiretroviral medication. The reasons for this are not known. It is also not known how many of these patients had taken antiretrovirals in the past and why these were no longer prescribed. Several variables were significantly associated with not being prescribed antiretroviral medication. These were: psychiatric morbidity as detected by the GHQ (odds ratio (OR) = 5.5, 95% CI 1.4–21.8, p < 0.01); scepticism about medication effectiveness (OR = 18, p < 0.001 by Fisher's exact test for small numbers (F)); and recent intravenous drug use (OR = 9.3, f = 0.013). On the other hand, patient age, CD4 cell count and a history of past psychotropic medication use were not significantly associated with prescription of antiretroviral medication.
Variables associated with adherence: multivariate analysis
There was significant interdependence between past psychotropic medication and current GHQ ‘case’ (OR = 5.6, 95% CI 2.1–15.0, p < 0.001), and between GHQ ‘case’ and scepticism about ARV medication (OR = 4.0, 95% CI 1.0–16.6, p < 0.05). All three factors that predicted poor adherence in the univariate analysis were highly interrelated. Because only four patients expressed scepticism about ARV effectiveness, all of whom exhibited poor medication adherence, logistic regression could not reliably determine the relative influence of medication attitude on adherence. Therefore, past psychotropic medication was not included in the final regression, due to intercorrelation with GHQ case, nor was attitude to medication included, due to low absolute numbers.
Logistic regression was performed, placing patient age in the first block of analysis, and GHQ case and illicit drug use in the second block. Of all other patient variables, only illicit drug use was weakly associated with poor adherence at p < 0.20 in the univariate analysis. GHQ case remained a significant predictor of poor adherence: OR was 4.1, 95% CI 1.3–12.6, p < 0.02. Neither age nor illicit drug use was significantly associated with adherence, p < 0.20 for both.
Discussion
These findings support the hypothesis that psychiatric morbidity, as measured on the GHQ, is significantly associated with poor adherence to antiretroviral medication. Most patients were male, reflecting the inner-city location and the principal mode of HIV transmission in urban Australian patients [29]. Most patients (84%) were taking antiretroviral medication. Three patient variables were found to be significantly associated with poor ARV medication adherence: past psychotropic medication, current elevated GHQ score and scepticism about effectiveness of medication. The variables were all closely interrelated.
Past psychotropic medication may be seen as a proxy for past psychological disturbance, which may reflect a predisposing trait, which is a forerunner of current psychological state, as detected by the GHQ. In turn, both anxiety and depression may predispose to pessimism and hence scepticism about the effectiveness of medication. GHQ ‘case’ remained a significant predictor of poor adherence using limited multivariate analysis, which included age and illicit drug use.
Questions about substance use failed to demonstrate a relationship between substance use and adherence. The questions were mostly open-ended and subjects may have given an underestimate of substance use. Very few patients reported recent intravenous drug use, so the numbers were too small to achieve significance. Disease knowledge did not correlate with adherence. In contrast to earlier thinking [11, 12], patients’ perceptions of current illness and future illness did not appear associated with adherence.
Subjects not prescribed medication experienced significantly more current psychological distress (GHQ case), expressed greater scepticism about medication and tended to use intravenous drugs, compared with those on antiretrovirals. It was not possible to determine why some patients were not prescribed antiretrovirals. Some may have decided to defer antiretrovirals while their viral loads were relatively low to avoid the sideeffects of medication. Other patients lead chaotic lives, with homelessness, substance misuse, difficult interpersonal relationships and a lack of stable routine, all contributing to inability to adhere to strict regimens. In such patients, the use of antiretrovirals may achieve only a modest benefit in viral load reduction but considerable harm in the form of developing viral resistance. Hence, their doctors may not offer antiretroviral treatment. Other patients may simply not wish to subject themselves to the demands of the regimen and its side-effects.
Two-thirds of patients reported good adherence to their medication. This is close to estimates of other authors, ranging from 60% to 67%, from the era of zidovudine [19, 20, 22], and the studies of Gordillo (58%) [6] and Haubrich (70%) [23] from the era of HAART. The rates are higher than the 38% reported by Bangsberg [7], who studied mostly IVDU and marginally housed patients, and 44% reported by Murphy [24], who interviewed only adolescents with HIV. The Clinic's patients were most similar to the patient groups described by Singh (1996) [20] and Haubrich [23], with HIV risk factors being mainly male-to-male sex and relatively low prevalence of IVDU.
Forty-four per cent of patients were identified as ‘cases’ on the GHQ-28. This is similar to Judd's finding of 50% depressed [30], below Gordillo's finding of 40% depressed plus 42% anxious [6], but higher than Lyketsos's 21% depressed [31] and Burack's 20% depressed [32]. However, our findings may represent an overestimate, as it is based on a GHQ cutting score of 4/5, as used in general practice patients [26], and may have been inflated by concurrent physical symptoms. Even when the cutting score was raised to 11/12, thereby reducing the number of ‘cases’, the association between GHQ case and poor adherence remained statistically significant.
There were several important reasons why medication adherence may have been overestimated. First, the study was cross-sectional, examining patients actually attending the clinic. It did not follow an inception cohort and could not assess those who dropped out from care. Irregular attenders may well have missed the 3 1 / 2 month enrolment period. Non-compliant patients tend to attend irregularly [10], tend to present late with symptoms, and are likely to miss out on audits of medication adherence. This source of bias is common to most compliance studies.
Second, it was not possible to account for patients who refused participation and those who either did not return the survey or returned it too incomplete to analyse. Such patients may tend to adhere poorly to medication. Third, patients were excluded because of poor English or lack of motivation to complete the questionnaire, and finally, the use of self-report is known to overestimate adherence, when compared with more objective measures such as the MEMS. Without comparing this particular self-report measure with a ‘gold standard’ such as the MEMS, it is difficult to know the extent of bias attributable to self-report of adherence.
Several studies have shown that medication adherence influences clinical outcome [3, 7, 23]. This study highlights the association of psychiatric morbidity with poor medication adherence, which in turn impacts on prognosis. It is not possible to infer causality because of the cross-sectional design of the study. As well, it is not known if treating patients with identified psychiatric disorder leads to improved ARV adherence and hence improved medical outcome.
The GHQ may be profitably used as a screening test for psychological disorder to augment clinical impressions [27, 28, 33]. Indeed, using the GHQ, Bell found psychiatric morbidity was often undetected among accident and emergency patients [33]. However, if HIV clinicians do not wish to administer such a questionnaire, then simply asking about past psychiatric history and previous psychotropic use may identify a large proportion of patients with psychiatric morbidity who are more likely to be poorly adherent to antiretroviral medication. This is suggested by the association found in this study between past psychotropic medication use and current GHQ-identified psychiatric morbidity.
This study has confirmed a significant association between current GHQ-detected psychiatric morbidity and poor medication adherence in an Australian outpatient clinic population. It is unknown whether psychiatric morbidity causes poor adherence and whether treatment of patients with identified psychiatric disorder leads in turn to improved adherence and better clinical outcome. A longitudinal study would have to be undertaken to investigate this.
Footnotes
Acknowledgements
We thank Roger Garsia, Kris Miller, Gary Trotter, Geoffrey Turnham and other staff of the Royal Prince Alfred Hospital's Immune Monitoring Clinic, Glenn Hunt for assistance with statistical analysis, and Neil Buhrich and Matthew Large for commenting on the manuscript.