Using Daily Interactive Voice Response Assessments

Abstract

Most clinical trials measure patient responses weekly, requiring patients to accurately recall and report their symptoms from the previous six days. More frequent assessments would be less susceptible to recall errors and recency effects as weekly assessments, but increased office visits burden clinicians and patients and can lead to higher attrition or non compliance. Interactive voice response (IVR) technology permits data collection at greater frequencies with minimal reporting burdens. An ancillary study within a randomized clinical trial evaluated the use of IVR to gather measures of patients' ratings of emotional and painful symptoms of depression on a daily basis. Unmedicated patients randomized to a starting dose of duloxetine of 30 mg QD (n=67) or 60 mg QD (n=70) called an IVR system daily to complete Verbal Numeric Scales for pain and Patient Global Impression of Improvement for both physical and emotional changes. Patients' compliance with daily IVR assessments was examined, and the IVR data obtained showed that patients started at 60 mg reported less pain and greater physical and emotional improvements than patients started at 30 mg. Dose related differences were evident as early as one day after the start treatment. This study provides new data about the usefulness of daily IVR assessments in clinical research and supports other studies regarding early symptom improvement with duloxetine.

Introduction

To measure the efficacy of treatments, most clinical trials use assessments requiring patients to retrospectively recall symptoms over a period of time (e.g., over the past week). Assessments administered on a weekly basis rely on patients' ability to accurately recall and report their symptoms over the previous six days. On the other hand, daily assessments are less prone to recall bias and recency effects, as they only require patients to report their symptoms on that given day. Comparisons of weekly and daily assessments of depression have found that patients' weekly reports of depression symptoms shared significantly more variance with the daily report obtained on the day the weekly assessment was conducted than with the average of the daily reports obtained over the previous six days.¹ In other words, when asked to report symptoms over the past week, patients tend to report how they are feeling on that day, and information regarding symptoms from the previous six days is lost.

A recent meta-analysis suggests that some antidepressants may improve symptoms of depression within the first week of treatment.² As most of the studies used weekly administered assessments, it is likely that symptom improvement occurred before the end of the first week of treatment. Unfortunately, assessments designed for weekly administration are inadequate for explaining clinical improvements that may occur within the first week. Reliably detecting onset of symptom improvement requires more frequent, repeated evaluations of patients' symtpoms.³ However, frequent office evaluations are impractical, costly, and burdensome. Studies using daily assessments to measure early improvements require remote assessment methods that are easily accessed and used by patients.

Interactive voice response (IVR) technology enables computer-based retrieval and processing of information through touch-tone telephones. IVR systems have been developed for a wide range of clinical applications, from screening for Axis I psychiatric disorders⁴ and dementia⁵ to daily monitoring of patients.⁶ Use of IVR systems to collect patient-reported outcomes (PROs) in clinical trials has become common.⁷ Using computer technology via telephones to collect PROs, IVR facilitates frequent, reliable and convenient collection of data while reducing the reporting burden for patients.⁸ Research regarding the feasibility of IVR daily assessments in clinical trials is of considerable current interest.

Daily IVR assessments of symptoms and clinical change were incorporated into a clinical trial as an ancillary methodological analysis from an industry-sponsored Phase IV investigation of duloxetine, a dual-reuptake inhibitor of serotonin and norepinephrine. Duloxetine has demonstrated efficacy in treating physical symptoms in patients with major depressive disorder (MDD).⁹ Previous studies found duloxetine demonstrated significant improvements compared to placebo on measures of depression, pain, and global functioning after one week of treatment.¹⁰ As these studies measured improvement using weekly assessments, it was not known if significant symptom improvement occurred before the end of the first week of treatment. In addition to examining the feasibility of daily IVR assessments, this ancillary study examined the potential utility of daily IVR assessments to measure symptom improvements in a compound with demonstrated efficacy. It was hypothesized that patients would complete daily IVR measurements and that such measures would detect symptom improvements within the first six days of starting duloxetine treatment, with patients on a higher dose reporting more improvement than patients on a lower dose.

Materials and Methods

The data obtained were part of a 12-week acute phase of an open-label, multicenter clinical trial of duloxetine (F1J-US-HMBZ) in the treatment of major depressive disorder (MDD).

Selection of patients. All participants were at least 18 years of age, and each signed an informed consent document prior to the administration of any study procedures or study drug. All participants met diagnostic criteria for MDD defined by the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV) and were not being treated with antidepressant medications at the start of the study. In addition, all enrolled patients had entry scores of Ž15 on the 17-item clinician-rated Hamilton Depression Rating Scale (HAMD-17)¹¹ and Ž4 on the Clinician Global Impression of Severity (CGI-S).¹² Patients were excluded from participation if they had any of the following: a diagnosis of bipolar disorder, schizophrenia, or other psychotic disorder; a primary and current Axis II disorder; serious medical illness; serious suicidal risk; a primary anxiety disorder diagnosis in the past six months; substance dependence in the past six months; or a positive drug screen.

Study design. The trial was conducted at 27 centers in the United States. The ethical review board at each site approved the protocol. During the first week of treatment, enrolled patients were randomized to a starting duloxetine dose of either 30mg QD (n=67) or 60mg QD (n=70) taken in the morning. The trial also included 112 patients who were switched from another antidepressant to duloxetine, but these patients are not included in the ancillary IVR methodology analysis. Concomitant medications with primarily central nervous system activity were not allowed. At the baseline office visit, patients completed paper Visual Analog Scales (VAS)¹³ for eight pain domains, and clinicians completed a HAMD-17, Hamilton Rating Scale for Anxiety (HAMA),¹⁴ and CGI-S for each patient. The VAS required patients to rate the severity of their recent pain symptoms by making a mark on a 100-mm line anchored by “no pain” at 0mm and “pain as bad as you can imagine” at 100mm. Between the baseline office visit and a follow-up visit scheduled one week later, patients were instructed to call an IVR system each day to complete assessments of pain and perceived improvement.

Daily IVR measures. Each day, patients completed two IVR assessments. First, Verbal Numeric Scales (VNS) gathered ratings of pain severity on eight domains using an 11-point scale ranging from 0 (“no pain at all”) through 10 (“pain as severe as you can imagine”). The domains included shoulder pain, back pain, headaches, pain while awake, abdominal pain, upset stomach, overall pain, and interference with daily activities. Second, the Patient Global Impression of Improvement (PGI-I)¹² assessed both physical and emotional changes by asking patients to rate change in their physical and emotional states since starting the study drug on a 7-point scale anchored by 1 (“very much better”) and 7 (“very much worse”), with a score of 4 meaning “no change.”

IVR system design. Patients accessed the IVR system via a toll-free telephone number given to them at the study site. The content of the system was recorded by a psychiatrist with experience in developing IVR clinical assessments (JHG). During their first call on the day after baseline, patients were required to verify that they had taken the study medication given to them at the office visit. Patients who reported that they had started taking their medication proceeded to the VNS and PGI-I assessments. Each assessment was preceded by a brief explanation of the questions that followed, and, for the VNS, patients heard a description of what was meant by “pain” and “discomfort.” The VNS questions were presented first, followed by the PGI-I Physical and PGI-I Emotional, respectively. Patients were instructed to respond to each of the questions by pressing a button on the telephone keypad. For each IVR call, an electronic record was created that contained a unique patient identification number, the date and time of the start and end of the call, and patient responses to the assessments and stored in an Oracle^® relational database. In addition to responses to the efficacy measures, these data enabled the examination of the number of calls received from the patients and the length and time of day of the calls.

Safety measures. Safety was assessed by means of spontaneously reported treatment-emergent adverse events, serious adverse events, discontinuations due to adverse events, and vital signs during the office visit following the first week of treatment. Results from the acute phase for this study for treatment-emergent adverse events and mean changes in vital signs are reported elsewhere.¹⁵

Statistical methods. Evaluation of ancillary results focused on the feasibility of assessing onset of treatment improvements and differences between the two groups during the first six days of treatment. All randomized patients with at least one post-baseline assessment were included in the analyses. Patient demographics were compared using the ANOVA model for continuous characteristics (age) and with Fisher's exact test for comparing percentages for categorical characteristics (gender, origin). Baseline scores for HAMD-17, HAMA, CGI-S, and VAS were compared using analysis of variance (ANOVA) including effects for group and investigator. Call duration compliance was also examined using ANOVA including effects for group and investigator.

The daily IVR measures were assessed using a likelihood-based, mixed-effects, repeated measures approach.¹⁶ The models included the fixed categorical effects of group and investigator. Time of assessment was modeled as a continuous effect by including linear and quadratic terms for days on therapy, as well as the interaction of the linear and quadratic terms with treatment group. Baseline severity measures were included as continuous covariates. Within-patient errors were modeled using an unstructured covariance matrix. If the unstructured covariance matrix failed to converge autoregressive, compound symmetric, and toeplitz structured covariance matrices, with and without heterogeneous variance by visit, were tested. The covariance structure converging to the best fit, as determined by Akaike's information criterion, was used.¹⁷ The Kenward-Roger method was used to estimate denominator degrees of freedom.¹⁸

Results

Patient characteristics. A total of 330 patients entered the screening phase of the study. Of these patients, 81 failed to meet entry criteria or declined to participate. The remaining 249 patients were comprised of those patients being switched from another antidepressant (n=112, not included in this analysis) and those included in this study who were randomly assigned to start duloxetine at 30mg QD (n=67) or 60mg QD (n=70). Demographics and baseline psychiatric information for the randomized patients are listed in Table 1. No significant differences were observed between treatment groups in demographics or baseline psychiatric information as measured by the HAMD-17, HAMA and the CGI-S.

Table 1.

Baseline Patient Demographics and Psychiatric Information

	30mg (n=67)	60mg (n=70)
Gender, n (%)
Female	38 (56.7)	44 (62.9)
Age, mean (SD)	42.27 (13.46)	41.98 (12.58)
Ethnicity, n (%)
African American	3 (4.5)	5 (7.1)
Caucasian	59 (88.1)	59 (84.3)
East Asian	0 (0.0)	2 (2.9)
Hispanic	5 (7.5)	4 (5.7)
Psychiatric profile, mean (SD)
HAMD-17	20.87 (3.73)	20.13 (3.25)
HAMA	16.88 (5.68)	16.71 (5.47)
CGI-S	4.36 (0.51)	4.29 (0.51)

Abbreviations

HAMD-17

Hamilton Depression Rating Scale

HAMA

Hamilton Anxiety Rating Scale

CGI-S

Clinical Global Impression of Severity

IVR calls. Throughout the six study days, patients completed 700 assessment calls. Compared to the 822 total possible calls (137 patients x 6 days), 85 percent of the expected calls were completed. Patients in the 30mg group provided 338 completed calls (48%), and patients in the 60mg group completed 362 calls (52%). Only three patients (2%) completed less than four calls. On average, patients completed 5.19 calls (SD=1.21) throughout the six days.

The IVR calls required an average of four minutes and 18 seconds to complete (SD=50 seconds). The shortest call lasted two minutes and 33 seconds, and the longest call lasted six minutes and 51 seconds. Eighty percent of the calls lasted less than five minutes. The duration of the calls decreased throughout the six days as users became more familiar with the system. The average call duration on Day 1 was 4:39 (SD=42 seconds), 4:24 on Day 2 (SD=51 seconds), 4:16 on Day 3 (SD=49 seconds), 4:11 on Day 4 (SD=50 seconds), 4:11 on Day 5 (SD=49 seconds) and 4:04 on Day 6 (59 seconds). The system could be accessed anytime, and calls were received during all times of the day. Fifty percent of the calls were received between 8:00 am and 5:00 pm; 35 percent after 5:00pm but before midnight, and 15 percent between midnight and 8:00 am.

Self-reported global improvement. The groups did not differ on the PGI-I ratings for physical improvements on Days 1 through 6. Figure 1 presents the adjusted means for the IVR reported PGI-I emotional ratings for each treatment group by day. The daily IVR reports by the 60mg group showed significantly greater emotional improvement by Day 3 compared to the 30mg group. Using Cohen's d (d=(M₁-M₂)/σ)¹⁹, the effect size for this differences was 0.40. The 60mg group continued to report significantly greater emotional improvement than the 30mg group throughout the remaining three study days. The effect sizes, reported as d, for the significant differences on Days 4 through 6 were 0.51, 0.68, and 0.81, respectively. Using Cohen's criteria,¹⁹ these effect sizes are considered medium to large.

Figure 1.

Adjusted mean PGI-I ratings for emotional improvement by treatment group and by study day

* p< .05

Abbreviations

PGI-I

Patients Global Impression of Improvement of Depression

Daily pain measures. Table 2 presents the means and standard deviations for each of the VAS pain domains at baseline and the adjusted means and standard deviations for the VNS pain domains on each of the six days by group. No significant differences between the two groups were found during the first six days of treatment on daily IVR ratings of back pain, stomach upset or interference with daily activities. Patients started at 60mg of duloxetine reported significantly less shoulder pain and headache by day one (d=0.38 and 0.38); pain while awake by day two (d=0.38), and overall pain by day four (d=0.36) compared to the 30 mg patients.

Table 2.

Baseline mean and standard deviation VAS and post-baseline adjusted mean and standard deviation VNRS scores by group and by day

		Baseline1		Day 1		Day 2		Day 3		Day 4		Day 5		Day 6
	n	30mg=67		30mg=56		30mg=57		30mg=58		30mg=57		30mg=49		30mg= 43
		60mg=66		60mg=55		60mg=55		60mg=56		60mg=53		60mg=50		60mg=45
Item	Group	Mean	SD	Mean	SD	Mean	SD	Mean	SD	Mean	SD	Mean	SD	Mean	SD
Shoulder pain	30mg	18.37	22.92	1.74	0.23	1.55	0.18	1.39	0.16	1.26	0.16	1.17	0.16	1.12	0.16
	60mg	11.2	20.24	1.10*	0.22	1.1	0.18	1.1	0.15	1.1	0.15	1.1	0.15	1.1	0.16
Back pain	30mg	27.09	27.19	1.98	0.23	1.85	0.2	1.73	0.19	1.61	0.19	1.49	0.19	1.37	0.19
	60mg	18.70*	23.42	1.41	0.22	1.35	0.2	1.3	0.18	1.26	0.18	1.23	0.18	1.22	0.18
Headaches	30mg	21.73	23.74	2.1	0.24	1.97	0.18	1.85	0.16	1.72	0.16	1.59	0.17	1.46	0.17
	60mg	22.95	25.68	1.43*	0.23	1.34**	0.17	1.26**	0.15	1.19*	0.14	1.13*	0.15	1.08	0.1
Pain while awake	30mg	38.19	33.68	3.78	0.34	3.7	0.3	3.62	0.3	3.54	0.3	3.46	0.3	3.38	0.31
	60mg	34.56	32.74	3.1	0.32	2.87*	0.29	2.68*	0.28	2.51**	0.28	2.38**	0.29	2.28**	0.3
Abdominal pain	30mg	7.07	12.3	1.43	0.21	1.26	0.17	1.13	0.15	1.03	0.14	0.97	0.14	0.94	0.14
	60mg	11.32	19.12	1.11	0.2	0.97	0.17	0.84	0.15	0.74	0.14	0.65	0.14	0.58	0.14
Upset stomach	30mg	14.3	19.86	2.4	0.31	2.03	0.22	1.72	0.17	1.45	0.15	1.24	0.14	1.08	0.13
	60mg	16.2	21.1	2.22	0.3	1.89	0.22	1.61	0.17	1.36	0.14	1.16	0.13	0.99	0.12
Overall pain	30mg	28.21	24.09	2.6	0.25	2.48	0.22	2.37	0.21	2.26	0.2	2.16	0.2	2.06	0.2
	60mg	25.14	22.09	2.14	0.24	1.98	0.21	1.84	0.2	1.73*	0.2	1.65	0.19	1.59	0.2
Daily activities	30mg	24.97	28.46	2.15	0.3	2.06	0.25	1.96	0.23	1.84	0.22	1.71	0.21	1.57	0.21
	60mg	20.67	25.72	1.83	0.28	1.68	0.24	1.54	0.22	1.43	0.21	1.33	0.21	1.24	0.21

¹Baseline pain ratings were gathering using a 100 mm VAS. Post-baseline ratings were gathered via IVR VNS (0-10 range); Post-baseline values were were adjusted for baseline differences

^*p<0.05 vs. 30mg

^**p<0.01 vs. 30mg

Abbreviations

VAS

Visual Analog Scale

VNRS

Verbal Numeric Rating Scale

The significantly less headache reported by patients in the 60mg group, starting at Day 1, persisted for Days 2 (d=0.48), 3 (d=0.51), 4 (d=0.47), and 5 (d=0.41). According to Cohen,¹⁹ these effects would be considered medium in size. For ratings of pain while awake, patients in the 60mg group provided ratings that were significantly lower than those for patients in the 30mg group at Days 2, 3 (d=0.42), 4 (d=0.48), 5 (d=0.52), and 6 (d=0.54). These effects start as small to medium and build to medium to large throughout the first six days.

Safety information. With respect to treatment-emergent adverse events, significantly more patients in the 60mg group reported nausea as a side effect than patients in the 30mg group (p=0.03). However, no significant differences were found for discontinuation due to nausea. As noted above, detailed safety results from this study are reported elsewhere.

Discussion

Until recently, researchers interested in measuring onset of treatment efficacy had few options other than weekly assessments asking patients to retrospectively report their experiences over the last seven days. This approach assumes that patients can accurately recall and report their symptoms across days within the referenced period. It also fails to capture improvements that may occur between the weekly assessments. Paper and pencil diaries have been used to collect data on a daily basis, but problems with retrospective and prospective data entry have been noted.²⁰ In contrast, daily IVR assessments include electronic time and date stamps that accurately identify the moment of data collection.

This ancillary study explored the feasibility of using daily IVR assessments to measure temporal changes in patient-reported symptoms in response to treatment in greater detail than current clinical trial methods permit. The results support the hypothesis that daily assessments to measure early symptom improvements are feasible. The results also are consistent with previous research suggesting that improvements with newer antidepressants may be seen within the first week of treatment,² rather than having more delayed effects as previously thought.²¹ In the current study, daily IVR assessments detected early symptom improvements that would not have been apparent or possibly not correctly recalled using standard weekly assessment methods. This study also provides evidence for convergent validity of the daily IVR assessments, as the obtained results are consistent with previous findings regarding early symptom improvement with duloxetine.⁹ Most patients in the study successfully completed the brief call each day, indicating that daily calling is not overly burdensome.

In addition to feasibility, the daily IVR assessments showed differences between the two groups of patients with respect to emotional improvements and pain. Given the relatively low levels of pain reported at baseline, that statistically significant differences in levels of reported pain were found is particularly striking. Previous research has established the superior efficacy of a 60mg versus a 30mg dose^22,23 but little was known about the onset of the improvements. Patients in the 60mg group reported significantly more emotional improvement after three days of treatment. A daily measure of depression measure was not obtained in this study but might have provided additional information about the nature of the more rapid emotional improvements experienced by patients in the 60mg group. For example, did certain depressive symptoms begin to improve before others? However, such a procedure also would have added to the patients' reporting burden.

In addition to supporting results of other studies showing that 60mg QD duloxetine is associated with significant reductions in physical symptoms,^9,24,25 the daily IVR assessments revealed two potential patterns of improvement in reported levels of pain. The effects for headache suggest that patients in the 60mg group reported immediate improvement, resulting in significant group differences. Over time, the 30mg group also reported fewer headaches, leading to a non-significant group difference at Day 6. Hence, patients in the 60mg group showed greater headache improvement upon starting treatment, but both groups of patients showed improvements after six days of treatment. In contrast, in terms of the results for pain while awake, patients in the 60mg group reported increasingly lower levels of pain while awake throughout the six days, resulting in a sustained significant difference from patients in the 30mg group, beginning at Day 2 and lasting through the entire first six days of treatment. The pain while awake reported by the 30mg group did not decrease to a level where the two means were no longer significant, as shown with headache. Daily symptom monitoring past Day 6 may have revealed additional patterns of symptom improvement but was not included in this feasibility study.

Several limitations of this study deserve comment. First, the sample sizes in this ancillary study were relatively small. Future research should explore the use of daily IVR assessments with larger samples. In addition, baseline reports of pain symptoms were collected via a paper-and-pencil VAS, while on Days 1 through 6 after baseline, patients completed an IVR VNS for pain. Although the correlation between these two measures is high (r=0.94),²⁶ ideally, an IVR VNS would have been administered at baseline. Finally, the results for the PGI-I for emotional change show significant differences between the groups starting at Day 3. Without a corresponding daily depression measure, no information is available regarding the specific symptom alleviation contributing to these emotional improvements.

Conclusions

IVR technology can provide a reliable method of gathering standardized patient-reported data at frequent intervals.⁸ The current investigation supported IVR assessment intervals as frequent as daily with good patient adherence. This procedure may result in more accurate and detailed data regarding patients' symptoms compared to the current practice of weekly assessments that rely on patients' ability to correctly recall and report their symptoms over a period of several days. This methodology may be particularly useful when studying onset of efficacy for compounds thought to provide rapid symptom relief.