A Mixed Model for Factors Predictive of Pain in AIDS Patients with Herpes Zoster

Transcription

1 410 Journal of Pain and Symptom Management Vol. 17 No. 6 June 1999 Original Article A Mixed Model for Factors Predictive of Pain in AIDS Patients with Herpes Zoster Renée A. Harrison, PhD, Seng-jaw Soong, PhD, Heidi L.Weiss, PhD, John W. Gnann Jr., MD, and Richard J. Whitley, MD Biostatistics Unit (R.A.H., S.-j.S., H.L.W.), Comprehensive Cancer Center, Department of Medicine (J.W.G.), and Departments of Pediatrics, Microbiology, and Medicine (R.J.W.), University of Alabama at Birmingham, Birmingham, Alabama, USA Abstract A unifying model of herpes zoster pain presents considerable analytical challenges due to the requirement for prospective data collection and the varying rates of pain resolution reported by individual patients. Demographic, clinical, and quality-of-life measures were collected on 166 human immunodeficiency virus (HIV)-infected patients enrolled in a randomized, controlled trial of antiviral therapy of herpes zoster comparing acyclovir with sorivudine. A mixed model was used to assess factors predictive of pain severity, activity impairment, and sleep interruption. The average rate of change in acute pain was 0.04 unit pain per day for the first month. Chronic pain decreased 0.12 per month for months Acute pain severity was positively correlated with number of new skin vesicles, analgesic use, and baseline pain, and negatively related to percentage of lesion healing and crusting. Postherpetic neuralgia was correlated with baseline pain, pain at 1 month, and duration of lesions. Treatment group, gender, race, and CD4 count were not related to change in pain severity. These analyses verify the significance of baseline pain as a significant predictor of pain resolution and average pain severity as a predictor of return to normal daily activities and sleep. The severity of acute pain at presentation and at 1 month are significant predictors of chronic pain. J Pain Symptom Manage 1999;17: U.S. Cancer Pain Relief Committee, Key Words Regression analysis, hierarchical model, epidemiological methods, pain Address reprint requests to: Renée A. Harrison, PhD, WTI 153, University of Alabama, 1824 Sixth Avenue South, Birmingham, AL , USA. Accepted for publication: September 17, Introduction Recent investigations have reported that demographic and clinical variables impact qualityof-life variables in patients with herpes zoster. 1,2 Commonly, the effects of therapy on pain and quality of life have been assessed at various predefined time points. Most analyses have utilized survival methodology which assesses activity limitations or time to cessation of pain between two (or more) groups. 3,4 However, acute herpes zoster pain and postherpetic neuralgia (PHN) represent distinct phases of a pain continuum. 5 It is of interest to explore how individuals differ in their rates of pain resolution and the factors that explain these differences. The statistical methods for analyzing longitudinal data have undergone recent improvements that now permit flexibility even when U.S. Cancer Pain Relief Committee, /98/$ see front matter Published by Elsevier, New York, New York PII S (99)

2 Vol. 17 No. 6 June 1999 Pain in AIDS Patients with Herpes Zoster 411 the data set is flawed by missing values. 6 8 Mixed models can be applied to analyze factors that change over time, such as pain, activity, analgesic use, and lesion healing, versus fixed factors such as race or treatment group assignment. The objective of this investigation is to describe the intraindividual and interindividual quality-of-life measurements among human immunodeficiency virus (HIV)-infected patients during the first month after onset of herpes zoster disease, and in the subsequent chronic period up to 12 months. Methods and Materials Patient Population This study, conducted by the National Institute of Allergy and Infectious Diseases (NIAID) Collaborative Antiviral Studies Group (CASG), enrolled 170 HIV-infected patients in a multicenter, randomized, double-blind, placebocontrolled clinical trial of the efficacy of acyclovir versus sorivudine for the treatment of herpes zoster. 9 HIV-seropositive patients at least 18 years of age who presented with a localized dermatomal vesicular rash of less than 72 hours duration were evaluated for enrollment. Patients were required to have a Karnofsky Performance Status Index of 60 (i.e., requires occasional assistance but is able to care for most personal needs ). All women of childbearing potential were required to have a negative serum pregnancy test. Excluded from the study were patients who were receiving topical or systemic therapy with other drugs active against varicella-zoster virus, were unable to take oral medication, had other acute or life-threatening opportunistic infections, had evidence of visceral or cutaneous dissemination of herpes zoster, and patients with laboratory evidence of significant renal, hepatic, or bone marrow dysfunction. Also excluded were patients receiving treatment with 5-fluorouracil (5-FU) or 5-FU derivatives because of the potential for a serious drug interaction with sorivudine. All patients gave informed consent before randomization. The informed consent document was approved by all local institutional review boards in compliance with regulations of the U. S. Food and Drug Administration. Data from 19 patients who had a recurrence of herpes zoster after the first month were included only up to the time of the recurrence. Two patients died during the acute treatment phase (at 12 and 13 days), and 16 patients during the 1 12 month period. Data from the latter 16 patients were included in the analyses up until the time of their death. All data were recorded on standardized record forms and analyzed at the CASG Central Unit, The University of Alabama at Birmingham. Clinical Variables Acute Phase ( 30 days). After enrollment, patients were evaluated daily until lesions were completely healed. Clinical variables assessed at each visit included the number of new vesicles, extent of healing (measured by presence of vesiculation, pustulation, crusting, and total healing), analgesic requirements, and body temperature. Duration of rash in days and baseline CD4 count ( 200 1, 200 0) were recorded at study entry. Long Term (1 12 months). After lesions were completely healed, patients were evaluated at monthly intervals up to 12 months for pain, analgesic use, level of activity, and sleep patterns. If pain was reported, the quality-of-life variables were assessed and analgesic requirements were assessed. Quality-of-Life Data Of seven different domains of pain-related functional interference, this study was able to measure activity and sleep. 10 A Karnofsky performance status score of 60 restricted the population to individuals with acceptable physical performance. No psychosocial measures were evaluated. Quality-of-life measurements were assessed daily. The specific scales included pain severity, activity impairment, and sleep interruption. Pain was self-assessed at baseline and over time on a five-point ordinal scale, with 0 no pain, 1 mild pain, 2 moderate pain, 3 severe pain, and 4 incapacitating pain. The reported pain was only that attributable to herpes zoster. A second baseline pain score for the chronic phase was assigned the value of selfreported pain on day 28 3 days, or if no pain score was available for this interval, the baseline was assigned the last pain score reported prior to day 25. Impairment of usual activity, as self-reported, was assessed on a four-point scale. Patients experiencing no impairment reported a score of

3 412 Harrison et al. Vol. 17 No. 6 June 1999 zero, 50% impairment of activities 1, 75% impairment of activities 2, and total incapacitation 3. Similarly, sleep interruption was assessed on a four-point scale with uninterrupted sleep receiving a score of 0, awakened 1 or 2 times with pain 1, awakened 3 6 times with pain 2, and virtual sleeplessness 3. Some response data were missing. At 3 weeks, responses were recorded for 87% of the patients, and the responses were equally distributed between the two treatment groups. At 6 months, 86% of the surviving patients had recorded pain scores, and at 12 months, 79% had recorded pain scores. Mixed Model The mixed model assumes a unique trajectory over time for each patient. Each subject is observed on at least two occasions, but the number and spacing of measurements need not be the same for all individuals. The error for each individual is assumed to be distributed independently and normally with a mean of zero and a constant variance, 2. For a short time duration series, when there are few observations per subject, the error structure can be assumed to be independent. The person level model assumes that the pattern of change varies within each individual as a function of patient-specific factors. When the frequency of measurement is relatively sparse (such as during the first month), a linear model is convenient. If more observations are available, more complex models can be investigated. The person level model is expressed as: Y it 0i 1i (day) it e it where Y it is the pain score for subject i at time t, i 1, , t 1,... 45; 0i is the expected pain score of subject i at time 0; 1it is the expected rate of change per unit time (day) in the pain score for subject i; and e it is the random within-subject error of prediction for subject i at time t. The between-person model views the patientspecific changes as varying across subjects, perhaps due to differences in patient factors such as race, gender, or treatment. This model is expressed as: 0i oo 0i and 1i 10 1i where oo is the grand mean pain score, 10 is the grand mean of rate of change of pain score, oi is the random effect of person i on pain score at baseline, and 1i is the random effect of person i on the rate of change in pain score. Results of modeling for fixed effects such as duration of rash and gender are considered significant at P A test of the hypothesis of interpatient homogeneity with respect to the baseline scores, or rates of changes in scores over time is computed as the 2 statistic. Variables were assessed independently for their relationship with the outcomes, and the results of these preliminary analyses were used to build the most parsimonious multivariate models. Results Descriptive The mean age of the patients was 36.6 years (range years), and 89.8% were male (Table 1). The majority of the patients were Caucasian (67.5 %); other racial or ethnic groups Table 1 Demographic and Clinical Characteristics of Patient Population Characteristic n(%) Treatment group Acyclovir 82 (49.4) Sorivudine 84 (50.6) Stage of HIV 1 12 (7.2) 2 58 (34.9) 3 16 (9.6) 4 80 (48.3) Gender Male 149 (89.8) Female 17 (10.2) Race White 112 (67.5) Black 28 (16.9) Asian 1 (0.6) Other 25 (15.0) Duration of rash at baseline 1 day 70 (42.2) 1 day 96 (57.8) CD4 count (55.8) (44.2) Baseline pain No pain 58 (34.9) Mild pain 33 (19.9) Moderate pain 43 (25.9) Severe pain 28 (16.9) Incapacitating pain 4 (2.4) Baseline activity impairment No impairment 85 (51.2) 50% impairment of activities 52 (31.3) 75% impairment of activities 22 (13.3) Total incapacitation 7 (4.2) Baseline sleep interruption Uninterrupted 70 (42.2) Awakened 1 or 2 times with pain 40 (24.1) Awakened 3 6 times with pain 33 (19.9) Virtual sleeplessness 23 (13.8) Age (mean SD)

4 Vol. 17 No. 6 June 1999 Pain in AIDS Patients with Herpes Zoster 413 included African American (16.9%), Asian (0.6%), and other (15.1%). Five female patients and 18 male patients experienced a recurrence of herpes zoster. Over half of the patients had a CD4 count of 200 at baseline. The majority of the patients (54.8%) reported mild or no pain at study enrollment. Of 170 patients enrolled in the trial, four patients had herpes zoster recur during the first month and were excluded from the analysis because the assessment of healing and evaluation of quality-of-life measures would be confounded; therefore, the number of patients evaluated for efficacy was 166. Acute Phase Figure 1 illustrates the resolution of pain severity by treatment group over the first month. There is an initial increase in pain from baseline to day 2, which is followed by a subsequent decrease over the ensuing 1-month period. Little difference in mean pain severity is evident between the two treatment groups. The models developed to fit analysis of pain severity over time are presented in Table 2. The first model indicates, as expected, that there are significant differences in the average pain measurements ( 0i 0.459, P ) as well as the rate of change of pain severity between patients ( 1i 0.009, P ). There is a decrease of 0.04 (95% confidence interval [CI] 0.037, 0.050) units of pain score per day. The second model adds baseline pain as a predictor, which is found to be significantly positively related to pain severity over time (P ). The third model introduces four time-varying covariates (analgesic use, extent of lesion healing and lesion crusting, and number of new vesicles), and two fixed patient factors (duration of rash and treatment group). There is a significant positive relationship between analgesic use and pain over time, as expected (P ). Table 3 shows the cross-tabulation of severity of self-reported pain and analgesic use during the first month. The number of new vesicles is also significantly positively related to pain scores over time (P ). In contrast, lesion crusting and healing are inversely related to pain severity scores. For a 10% increase in lesion crusting at a given time, the pain score decreases by 0.05; for 10% increase in lesion healing, the pain severity decreases by Although duration of rash at baseline is positively related to pain, the P value is not signifi- Fig. 1. Mean pain severity in first month among HIV patients with herpes zoster in randomized drug trial.

5 414 Harrison et al. Vol. 17 No. 6 June 1999 Table 2 Regression Coefficients and 95% CI for Mixed Models Predicting Pain Severity in the First Month a Time only Coefficient (95% CI) Time plus baseline pain Coefficient (95% CI) Time and all covariates Coefficient (95% CI) Fixed covariates Model for pain means Intercept (1.702, 1.939) b (1.464, 1.770) b (0.970, 1.326) b Base pain score (0.075, 0.224) b (0.046, 0.156) b Duration of rash 0 1 day ( 0.007, 0.261) 1 1 day Group Acyclovir ( 0.056, 0.208) Sorivudine 1 Time (days) ( 0.037, 0.050) b ( 0.036, 0.050) b ( 0.025, 0.010) b Analgesic use (1 5) (0.332, 0.433) b Healing (%) ( 0.005, 0.002) b Crusting (%) ( 0.006, 0.003) b New vesicles (number) (0.001, 0.004) b a Only 158 patients had sufficient data to be included in model computations. b P cant (P 0.06). Although patients differed significantly in rate of change of pain, no factors tested were significantly associated with changes in pain severity, including CD4 count (P 0.097), gender (P 0.984), age (P 0.844), temperature (P 0.933), and unhealed surface area (P 0.640). Finally, treatment group is not a significant interpatient factor (P 0.26), as has been previously reported. 9,11 For the time-dependent outcomes of return to usual activity and ability to sleep uninterrupted, scores at day 1 were considered baseline levels. Final model results are shown in Table 4. Because of the correlations between activity level and pain severity (r 0.56), and sleeping indices and pain (r 0.58), once the pain score was introduced into the model, there was no significant time effect. That is, controlling for pain severity over time, the slope of activity impairment or sleep index was not significantly different than zero (P and P 0.429, respectively). Therefore, the models presented in Table 4 include the same covariates as in Table 2, except for analgesic use. In both models, baseline levels were significantly positively associated with average impairment scores (P ), and lesion crusting and healing were related inversely to impairment average scores. There was no significant effect of treatment group assignment for either impairment of activity or inability to sleep uninterrupted. Long-term Follow-up A model fit to pain evaluated over time is shown in Table 5. For this model, the measurements within the first month were assigned time values to correspond with months (e.g., divide time in days by 30); thereafter, only monthly assessments were recorded for each Table 3 Cross-tabulation of Severity of Pain with Use of Analgesic Agents during the First Month a No analgesics Non-opioid or antidepressant Oral opioid Oral opioid and antidepressant Parenteral opioid Total No pain Mild pain Moderate pain Severe pain Incapacitating pain a Values are percentage of all observations for each pain category.

6 Vol. 17 No. 6 June 1999 Pain in AIDS Patients with Herpes Zoster 415 Table 4 Regression Coefficients and 95% CI for Mixed Models Predicting Activity Impairment and Sleep Index in the First Month a Activity impairment Coefficient (95% CI) Sleep Index Coefficient (95% CI) Fixed covariates Model for means Intercept (0.462, 0.726) b (0.489, 0.814) b Base score (0.173, 0.297) b (0.108, 0.233) b Duration of rash 0 1 day ( 0.056, 0.152) ( 0.050, 0.223) 1 1 day Group Acyclovir ( 0.037, 0.169) ( 0.070, 0.198) Sorivudine 1 Time (days) ( 0.016, 0.005) b 0.007, ( 0.015, ) Healing (%) ( 0.004, 0.003) b 0.004, ( 0.005, 0.004) b Crusting (%) ( 0.004, 0.002) b 0.003, ( 0.005, 0.001) b New vesicles (number) ( , 0.002) (0.0008, 0.005) b a Only 158 patients had sufficient data to be included in model computations. b P patient. The expected pain severity for any patient at a time point was significantly positively related to baseline pain, duration of rash, and pain severity at 1 month. Treatment group was not significantly related to pain severity. The rate of change in pain severity varied significantly between patients, and was significantly negatively related to months since disease onset (P 0.001). The change in pain severity was 0.12 ( 95% CI 0.14, 0.10) for each month. The rate of change in pain was also significantly negatively related to baseline pain (P 0.044) and pain after 1 month (P 0.013). That is, the greater the baseline pain and pain after 1 month, the greater the decrease in pain severity at any given point in time. The negative covariance between the intercepts and slopes (not shown) implied that with higher baseline severity of pain, the patients showed larger reductions in pain over time, as expected. As in the previous models, there was no significant effect of gender, age, CD4 count, race, or treatment group on pain severity. The linear model for the cumulative pain data did not provide the best data fit. A plot of the mean pain severity over time by group as shown in Fig. 2 indicated a curvilinear pattern. Early in the disease phase, the severity of pain decreased more rapidly than in the subsequent months. The rate of reduction in pain did not differ significantly by treatment group. Figure 2 suggested a piece-wise exponential function with a starting value at baseline, and a second starting point between 1 3 months. Discussion These analyses identified three clinical variables (extent of lesion healing, extent of lesion crusting, and number of new vesicles), and two disease-related variables (baseline pain and analgesic use) which significantly predicted resolution of acute pain among this cohort of HIV- Table 5 Regression Coefficients and 95% CI for Mixed Model Predicting Pain Severity from Months 1 12 a Coefficient (95% CI) Fixed covariates Model for pain means Intercept (0.685, 0.943) b Baseline pain score (0.093, 0.207) b Baseline pain at one month (0.271, 0.435) b Duration of rash 0 1 day (0.064, 0.238) b 1 1 day Group Acyclovir 0 Sorivudine ( 0.093, 0.080) Model for pain slopes Intercept (months) ( 0.139, 0.104) b Baseline pain ( 0.019, 0.002) b Baseline pain at one month ( 0.031, 0.004) b a Only 165 patients had sufficient data to be included in computations. b P 0.05.

7 416 Harrison et al. Vol. 17 No. 6 June 1999 Fig. 2. Mean pain severity over 1 12 months among HIV patients with herpes zoster in randomized drug trial. infected patients treated for herpes zoster. Interpatient factors, including treatment group, race, sex, age, CD4 count, and duration of rash, were not significant. The resolution of pain over 1 month correlated with baseline pain, duration of rash, and pain score at 1 month. In addition, the rate of change in pain score from months 1 12 was negatively related to baseline pain and pain after 1 month. It has been reported that the strongest risk factor for PHN is the severity of the herpes zoster pain at presentation. 12 This observation was substantiated in the current study. Patients with greater baseline pain and more pain at 1 month have greater persistent pain scores. In a randomized trial of immunocompetent patients with herpes zoster, Whitley et al. 4 reported that severity of pain at baseline and number of lesions before enrollment predicted time to resolution of pain return to uninterrupted sleep and return to usual activity. The current report did not find that age or gender were related to pain severity or rate of change in pain severity. When variability of age is restricted (75% of the patients were age 39 or younger in the current study), age may not be a significant predictor of pain. Herpes zoster may impact quality of life through associated pain that affects daily activities, including sleeping. Although pain is lessened by use of analgesics, relationships between severity of pain and both activity and ability to sleep uninterrupted are apparent. The relationships between these time-varying factors that may be endogenous variables should be interpreted cautiously. For example, the effect of analgesic use on pain is likely to be reciprocal, i.e., patients with greater pain report greater analgesic use. Unfortunately, we were not able to assess psychosocial dimensions as related to herpes zoster pain over the first month. Some investigators have reported a relationship between patients who develop PHN and depression, anxiety, and lower life satisfaction. 12 However, pain and social support describe only 25% of the variation in overall psychological wellbeing of patients with AIDS. 13 It would be of interest to capture the psychosocial dimension of pain-related functional interference in future randomized clinical trials of herpes zoster. At the present time, algorithms to model nonlinear repeated measures data are the subject of active research. Clinical information about the pathophysiology of herpes zoster pain will aid in the design of statistical approaches for analysis of longitudinal data, particularly when the change in pain score is time-

8 Vol. 17 No. 6 June 1999 Pain in AIDS Patients with Herpes Zoster 417 dependent and differs between individuals. As suggested, most patients report pain as a continuum, but the pathophysiologic mechanisms of acute pain and postherpetic neuralgia may well be different. 3,12 However, current debate exists over how to delineate phases of pain. Furthermore, the time points for transitions between pain phases are likely to differ among individuals. One problem noted earlier with estimations of the rates of change in quality-of-life scores over time was the number of missing data points. Our estimates were more reliable for the intercept (pain severity at any given time) than the slopes (rate of pain change over time). If a more complete data set over time was available, the reliability of the estimates of the slope coefficients would be greater. Our analyses assumed that missing data was missing randomly, but patients with incomplete data are included. If the patient attrition is correlated with the outcome (pain), there may be bias in the estimation of effects. Because the factors predictive of pain are largely unknown, we cannot determine what effect our assumption has on this analysis. In conclusion, evaluation of postherpetic pain and the relationship to measured qualityof-life outcomes is an important area for future research. Herpes zoster pain has been noted to influence physical functioning, emotional functioning, and vitality. 2 Data should be collected prospectively on the domains of physical activity and emotional functioning in clinical trials. Future studies should also consider standardization of quality-of-life assessments so that efficacy studies may be compared across patient populations. Enrollment in clinical trials should target patients at highest risk for postherpetic neuralgia. 14 Recent investigations have suggested that significant gains in quality of life might be achieved with treatments that lead to more rapid cessation of zoster pain. 1 Acknowledgment The authors thank Ms. Linda Austin for assistance with data management. References 1. Mauskopf JA, Austin R, Dix LP, Berzon RA. Estimating the value of a generic quality-of-life measure. Med Care 1995;33(4S):AS195 AS Lydick E, Epstein RS, Himmelberger D, White CJ. Herpes zoster and quality of life: a self-limited disease with severe impact. Neurology 1995;45(S): S Wood MJ. How to measure and reduce the burden of zoster-associated pain. Scan J Infect Dis (suppl) 1996;100: Whitley RJ, Weiss H, Gnann JW, Tyring S, Mertz GJ, Pappas PG, et al. Acyclovir with and without prednisone for the treatment of herpes zoster. Ann Intern Med 1996;125: Portenoy RK, Duma C, Foley KM. Acute herpetic and postherpetic neuralgia: clinical review and current management. Ann Neurol 1986;20: Laird NM, Ware JH. Random-effects models for longitudinal data. Biometrics 1982;38: Beacon HJ, Thompson SG. Multi-level models for repeated measurement data: applications to quality of life data in clinical trials. Stat Med 1996; 15: Bryk AS, Raudenbush SW. Hierarchical linear models: applications and data analysis methods. Newbury Park, CA: Sage Publications, Gnann JW Jr, Crumpacker S, Lalezari JP, Smith JA, Tyring SK, Baum KF, et al. Sorivudine versus acyclovir for dermatomal herpes zoster in HIV-infected patients: results from a randomized, controlled clinical trial. Antimicrobial Agent Chemo 1998;42: Breitbart W, McDonald MV, Rosenfeld B, Passik SD, Hewitt D, Thaler H, Portenoy RK. Pain in ambulatory AIDS patients. I: Pain characteristics and medical correlates. Pain 1996;68: Bodsworth NJ, Boag F, Burdge D, Généreux M, Borleffs JCC, Evans BA, et al. Evaluation of sorivudine (BV-araU) versus acyclovir in the treatment of acute localized herpes zoster in human immunodeficiency virus-infected adults. J Infect Dis 1997;176: Dworkin RH, Portenoy RK. Pain and its persistence in herpes zoster. Pain 1996;67: Rosenfeld B, Breitbart W, McDonald MV, Passik SD, Thaler H, Portenoy RK. Pain in ambulatory AIDS patients. II: Impact of pain on psychological functioning and quality of life. Pain 1996;68: Kost RG, Straus SE. Drug therapy: postherpetic neuralgia pathogenesis, treatment and prevention. NEJM 1996;335:32 42.