Top, Relationship between change in symptom distress during drug treatment of angina pectoris and accompanying change in quality of life in units of SD. Note the concordance between the subscales of each cohort. Bottom, The relationships were essentially identical in a study in which hypertension treatment was the goal.
Hollenberg NK, Williams GH, Anderson R. Medical Therapy, Symptoms, and the Distress They CauseRelation to Quality of Life in Patients With Angina Pectoris and/or Hypertension. Arch Intern Med. 2000;160(10):1477-1483. doi:10.1001/archinte.160.10.1477
Adverse events during drug therapy can be assessed through measurement of 2 features: their frequency and their severity. Their severity, in turn, can be measured by assessing the distress that they cause. Our goal was to relate the magnitude of the distress induced by treatment with calcium-channel blocking agents to the change in quality of life assessed through psychosocial instruments in patients treated with calcium-channel blocking agents, either for hypertension or for angina pectoris.
Four hundred seventy-five patients with angina pectoris were randomized to double-blind treatment with PPR (physiological pattern release) verapamil hydrochloride, amlodipine besylate, amlodipine-atenolol combination, or placebo. In addition, 557 hypertensive patients were randomized either to PPR verapamil or nifedipine GITS (gastrointestinal system). Both studies were double-blind.
Significant differences in treatment of angina pectoris or hypertension, were not found between the regimens. Overall quality of life also failed to show a significant difference in either group. In both groups, however, remarkable concordance was found between the degree of distress associated with specific symptoms and a change in quality of life. An unchanged, stable symptom distress was associated with a significant improvement in the quality of life of about 0.1 SD. Improvement or erosion of symptom distress represented by 1 step was associated with a 0.1- to 0.2-SD change. The extreme change in symptom distress was associated with a substantially larger change in global quality of life.
The magnitude of symptom distress or relief associated with symptoms in 2 patient populations correlated strongly with a shift in quality of life. The assessment of distress associated with symptoms provides valuable additional information on drug therapy.
ONE OF the fundamental tenets of medicine is the Hippocratic injunction that we avoid doing harm. In the case of medical therapy, the most common harm that we wish to avoid is symptoms that result from our choice of treatment. When adverse events are sufficiently severe, they lead to discontinuation of treatment, with consequent loss of the benefits that treatment might provide.1,2 Clinical trials predict inadequately withdrawal from treatment in clinical practice.3 Thus, attempts to quantify the frequency of symptoms induced by treatment, and their severity, are fundamental in therapeutics.4,5
The instruments used to capture and evaluate information on symptom distress, as is the case for the instruments used for the assessment of quality of life (QOL) in general, require "calibration." How much of a change in QOL as measured by various instruments reflects what magnitude of shift in a person's life? Investigators have sought to address that question in several ways. The change in QOL associated with a decision to withdraw from medical therapy provided a compelling index.6 In an alternative approach, the change in QOL associated with spontaneous life events was assessed prospectively: the outcome related changes in QOL assessed by the instruments used in this study to easily understood standards such as loss of a job, divorce, or death of a spouse.7
During a study on symptoms and the distress associated with them in a therapeutic trial comparing calcium-channel blocking agents, we made an unanticipated observation: There was a high degree of correlation between symptom distress and changes in QOL as assessed by an instrument package that we had already calibrated.8 As we had no a priori hypothesis on this relationship in that study, the relationship was not developed in our original report. In this report, we describe that relationship, and a second assessment made prospectively in a second study involving a different patient population.9 In this study, we obtained a fresh database and compared prospectively the findings with the original association recognized in the earlier study. The fact that 2 groups of patients, one with essential hypertension and the other with angina pectoris, have shown an identical relationship between symptom distress changes and QOL suggests that a fundamental relationship is involved, which is the focus of this report.
The primary objective of the 2 studies was to evaluate the efficacy and safety of PPR (physiological pattern release) verapamil hydrochloride compared with amlodipine besylate, used alone or in combination with atenolol in patients with chronic stable angina pectoris9 or with nifedipine GITS (gastrointestinal system) in patients with hypertension.8 The clinical outcomes of both studies have been presented.8,9 This analysis deals only with changes in QOL in relation to symptom distress or relief.
A sufficient number of patients with angina pectoris was enrolled to ensure that a total of 475 would be randomized into the study. Of 832 patients who were screened at 79 centers, 475 were randomized into the double-blind treatment phase. Randomization assigned 153 participants to PPR verapamil, 128 to monotherapy with amlodipine, 128 to the amlodipine-atenolol combination, and 66 to placebo. In addition to a history of chronic stable angina pectoris without significant change in frequency or severity during the 3 months before screening, entry criteria included an age range of 30 to 80 years, ability to walk on a treadmill to perform an exercise tolerance test, occurrence of at least moderate angina or anginal equivalent between 2.5 and 8 minutes after the initiation of a standard Bruce exercise tolerance test, and evidence of a coronary ischemia from the exercise tolerance test.
In the hypertension study, 557 patients were randomized to either PPR verapamil (276 patients) or nifedipine GITS treatment (281 patients), as described in detail.8 The entry criteria in this study involved a seated systolic blood pressure without treatment between 140 and 190 mm Hg or a seated diastolic pressure between 90 and 115 mm Hg. After a placebo run-in, patients were randomized to treatment with either PPR verapamil or sustained-release nifedipine.
In the angina study, a 2- to 4-week single-blind placebo lead-in involved return to the clinic every 7 ± 2 days for 4 baseline visits.9 The patients were then randomized to active double-blind treatment for 4 weeks.
During that interval, they had office visits at the end of week 2 and at the end of week 4. On completion of study participation, the patients received treatment for their angina in accord with the clinical judgment of the responsible physician.
In the hypertension study, the protocol sequence differed in that active, double-blind treatment was maintained for 12 weeks rather than 4 weeks, and that visits for QOL assessment were made at 5 intervals, rather than 4, as described in detail.8
Procedures and instruments used in this study were similar to those previously used.6- 8 At each study center, a specifically trained office employee proctored the battery of patient self-administered questionnaires. The battery included assessment of several psychosocial measures (anxiety, loss of behavioral or emotional control, depression, "emotional ties," general positive affect, general health status, and vitality) and a physical symptom distress index. This index consisted of 73 items, including 2 that were not used in the present analysis because they applied only to men. The questionnaire presented a list of symptoms associated with hypertension, angina pectoris, or side effects from all of the drugs used. Each question asked the respondent, "Have you had this symptom at all in the past month?" The respondent was to circle a number between 0 (didn't have it at all) and 5 (yes, and it has bothered me extremely).
The psychological scales included the 5 subscales of anxiety, loss of behavioral or emotional control, depression, emotional ties, and general positive affect from the Rand Mental Health Index (MHI). Seven other items outside the MHI published by Dupuy10 provided 2 additional subscales (general health and vitality).
In the angina study, QOL assessments were made at entry, at the end of the baseline placebo run-in interval, and at 2 weeks and 4 weeks of active study with a randomized drug. In the hypertension study, the battery of instruments was administered 5 times: at the time of initial enrollment in the study, to acclimate the subject to the testing procedure; at the end of the placebo washout; at the completion of the study or at the time of dropout; and at 4 and 8 weeks after starting treatment.
To assess the global QOL at baseline, we used "Cantril's ladder."11 In this global QOL instrument, a visual analog scale is used in the vertical, rather than horizontal, orientation. A 10-rung ladder replaces the usual response line. The participant is asked to grade their current QOL on the ladder. Standing on the ground (0) represents the worst QOL that one could imagine. Standing on the top rung of the ladder(10) represents the best QOL that can be imagined, and can only be sustained for a short time. Standing on the middle rung (5) represents "the average QOL in your community," without further definition. Because baseline QOL can act as a determinant of QOL during treatment,7 we administered Cantril's ladder at baseline to provide an alternative to the instruments used for therapeutic evaluation to block for baseline effects.
Sample sizes in both trials were designed to accommodate the needs of the main efficacy assessment. In the angina study, the assessment was based on the change from baseline in symptom-limited exercise measured via the exercise tolerance test. The goal was to provide approximately 90% power to detect a true mean difference of 0.64 minute (38.4 seconds) between placebo and each of the 3 active treatment groups without adjustment for a multiplicity of testing. In the case of the hypertension study, sample size was predicated on the provision of at least 90% power to detect a between-group difference of 3 and 5 mm Hg in diastolic and systolic blood pressure, respectively.
The efficacy analysis was conducted on an intention-to-treat basis, with missing values being handled by means of the "last observation carried forward" method. Efficacy analysis was conducted by means of analysis of variance with baseline value, treatment, and center included in the model. Sample size was not computed to ensure any predetermined level of power in the QOL analysis.
The QOL analysis was conducted on an intention-to-treat basis. To be included in the analysis, the patient had to have a valid baseline and end-point QOL assessment. Treatment effects on baseline to end-point change in QOL were measured by 2 multivariate 1-way analyses of variance (MANOVAs), one for physical symptom distress and one for the psychosocial index. Pillai's trace statistic was used as an exact test of the multivariate significance in each analysis.12 If a main MANOVA detected an overall multivariate effect, the associated univariate effects were analyzed. The foregoing MANOVA also was repeated as a 2-way factorial analysis of treatment and dropout on symptom distress and psychosocial well-being. In addition to performing a MANOVA on the change score, a similar MANOVA was performed on the QOL questionnaires at the conclusion of the study in a cross-sectional analysis. All QOL analyses were implemented in SPSS.13
Table 1, Table 2, Table 3, and Table 4 compare the distribution of the 4 randomized treatment groups in the angina study with respect to demographic and baseline characteristics. They show no significant differences in the breakdown by sex, ethnicity, employment status, initial QOL, or baseline angina status. A similar baseline homogeneity was described in the hypertension study.8
During active treatment, premature discontinuation was uncommon, and the rate did not differ by treatment (18 [11.8%] in the verapamil group, 5 [3.9%] for amlodipine, 14 [10.9%] for amlodipine-atenolol, and 4 [6.1%] for placebo). Again, the findings in the hypertension study on withdrawal from treatment were similar.8
Under MANOVA, the 7 psychosocial subscales showed no significant multivariate treatment effect (P=.39) in the angina study. The psychosocial scales also showed no treatment-related effect under multivariate analysis in the hypertension study.8
In the angina study, the 1-way MANOVA showed a significant overall multivariate difference between the treatments (P=.006). Univariate analysis subsequently demonstrated significant 6-treatment effects in the angina study, all of borderline significance. The univariate differences among the treatment groups are listed in Table 5.
There were 5 possible distress level transitions: none to extreme; 1 step worse; no change; 1 step better; and extreme to none. In Figure 1, changes in the Rand MHI in SDs are described in relation to changes in reported symptom distress for the angina study (top) and for the hypertension study (bottom). For each change in MHI, the data are presented in units of the unadjusted sample SD of the index since the preceding visit. The size of the confidence interval reflects the number of individual values, which was largest in the "no change" category, was smaller but still substantial in the 1-step-better and 1-step-worse categories, and ranged from only 100 to 200 cases in the none-to-extreme and extreme-to-none categories. An identical set of relationships is evident in both studies. In both studies, no change in reported symptom distress was associated with a significant improvement in the MHI of about 0.1 SD. In both studies, improvement or erosion of symptom distress represented by 1 step was associated with a change of 0.1 to 0.2 SD. The extreme change in symptom distress in each study was associated with a substantially larger change in the MHI.
Intuitively, one would expect the highly correlated subscales in the MHI to be associated with symptom distress in much the same way as their composite. As also shown in Figure 1, that anticipation was reflected in the findings. In the no-change group in both studies, all 5 subscales showed a positive shift. In both the 1-step-better and 1-step-worse categories and in the extreme changes, there was complete accord within the 5 subscales. Emotional ties appeared to change least, especially at the extremes, but no other pattern was discernible.
The primary efficacy outcome was the subject of previous detailed reports.8,9 In general, all treatments improved angina pectoris. The QOL and symptom distress relationships described in the 2 studies were independent of relief of angina pectoris or blood pressure control.
The primary hypothesis underlying this report involved a possible relationship between changes in the magnitude of distress associated with a number of physical symptoms and changes in QOL as assessed by the Rand MHI.14 This relationship was noted, but not developed, in one study8 and became the primary hypothesis in a follow-up analogous study. The disease under treatment was hypertension in the first study and angina pectoris in the second. The remarkable accord in the relationships identified in the 2 studies indicates that the cardiovascular process under treatment made little or no contribution to the relationship. We had argued that antihypertensive therapy can erode QOL through 2 mechanisms: a direct effect on mood, as in the case of central nervous system–active agents and β-blockers, and an indirect action based on physical symptoms, as in the case of calcium-channel blocking agents.8 The range of symptoms identified in these studies, including ankle swelling, headache, constipation, facial flushing, and awareness of one's heartbeat, are those widely associated with calcium-channel blocking agents and come as no surprise.
In an earlier study in which we attempted to calibrate QOL instruments, including the MHI, through the prospective and systematic assessment of life events, we made an observation immediately relevant to the findings in this study.7 The threshold change in QOL that appeared to be clinically significant was evidenced by a change in measured QOL of about 0.1 SD—the same measure as the change assessed with a single step in physical symptom distress defined in this study.
In our earlier study, this level of erosion of QOL approximated that which occurred with concern about one's job. Larger changes, from 0.2 to 0.4 SD, were associated with larger and more important life event changes, such as loss of one's employment, divorce, or death of a spouse. Whether the none-to-extreme change in physical symptom distress is associated with a QOL change that truly matches the impact on one's QOL of divorce or death of a spouse cannot be answered from the data in this study. Given the language in the instruments, it would come as no surprise that they lack precision at the extreme. Another new observation in this study was the relative symmetry between improvement and erosion of QOL. In our earlier study7 that used life events as the index, there were too few positive events with which to assess improvement in QOL, although there were an ample number for the assessment of erosion of QOL. In this study, improvement and increase in symptoms occurred symmetrically, and improvement in QOL followed symmetrically. These are the first data in the QOL area, we believe, to demonstrate that symmetry.
One implication for these findings involves power calculations for future studies. In our earlier study, we pointed out how many studies in this field were underpowered, from the perspective of a 0.1- to 0.2-SD change.7 At the conventional level of 0.90 confidence and a commonly chosen level of 0.80 power, the sample size required for a single comparison of means would be 199 respondents divided equally between 2 randomly drawn samples from a large population. If one requires balance, confidence, and power each at 0.95, the required sample size increases to 325.
The improvement in QOL as assessed by the MHI on psychosocial characteristics in patients in whom there was no change in physical symptom distress is intriguing and currently unexplained. Although it was not associated with treatment, not differing during the placebo or active treatment phase, it might reflect a placebo effect. An alternative possibility is familiarity with the instruments, or increasing optimism as a study evolves. The incorporation of a large parallel placebo group in future studies should help to clarify this issue.
Although it is widely recognized that the subscales in the MHI are highly correlated, the remarkable concordance in the pattern of response was not anticipated. With each step of shift in symptom distress, all 5 subscales in the MHI showed a directionally identical and quantitatively similar shift. Indexes of anxiety and depression were the most robust, and emotional ties showed the least change. Presumably, either physical symptoms have a larger influence on one's mood than on one's feeling of being loved or emotional ties are more difficult to measure.
Assessment of the frequency of symptoms is a standard in therapeutics, but assessment of the severity of a symptom is much less common. Another outcome of this study is that assessment of severity, at least as performed in this study, carries substantial information. In the present studies, traditional psychosocial QOL indexes showed no treatment effect. Although the term validity is widely used in the psychosocial area, much of the validation is statistical or technical, reflecting primarily internal consistency, and does not ascertain whether an instrument is adequate to the task.15- 18 The instruments used in the present study have been validated by application in a series of investigations. In a comparison of captopril, propranolol hydrochloride, and methyldopa, a similar package of instruments showed a QOL profile that paralleled withdrawal rates and symptom patterns.19 In a comparison of atenolol with nifedipine, the QOL assessment predicted the dropout rate.6 We made a similar finding in the earlier study comparing PPR verapamil and nifedipine GITS.8 In a comparison of captopril and enalapril, a strong correlation was shown between QOL changes with treatment and the simultaneous occurrence of spontaneous life events coded prospectively.7 This study provides a strong confirmation and extends these observations to the domain of physical symptoms. The threshold for a shift in physical symptoms, as in the case of psychosocial indexes, was 0.1 SD. By the use of a full range of instruments, the physical symptom distress index was significantly related to treatment and predicted dropout rate.8
We conclude that the physical symptom distress index is a sensitive technique for assessing the influence of cardiovascular treatment on QOL. In studying treatment effects on QOL, both the distress specifically associated with physical symptoms and the change in psychosocial factors should be evaluated. The distress associated with specific symptoms escapes the ambiguity that is implicit in the assessment of QOL.20
Accepted for publication September 22, 1999.
This study was supported by a grant from G. D. Searle & Co, Skokie, Ill.
Reprints: Norman K. Hollenberg, MD, PhD, Brigham and Women's Hospital, 75 Francis St, Boston, MA 02115 (Fax: 617-232-2869).