This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of Medical Internet Research, is properly cited. The complete bibliographic information, a link to the original publication on http://www.jmir.org/, as well as this copyright and license information must be included.
Previous randomized controlled trials (RCTs) have shown a significant intervention effect of internet-based computerized cognitive behavioral therapy (iCBT) on improving nonclinical depressive symptoms among healthy workers and community residents in a primary prevention setting. Time preference is one’s relative valuation for having a reward (eg, money) at present than at a later date. Time preference may affect the effectiveness of cognitive behavioral therapy.
This RCT aimed to test the difference of intervention effect of an iCBT program on improving nonclinical depressive symptoms between two subgroups classified post-hoc on the basis of time preference among workers in Japan.
All workers in one corporate group (approximate n=20,000) were recruited. Participants who fulfilled the inclusion criteria were randomly allocated to either intervention or control groups. Participants in the intervention group completed 6 weekly lessons and homework assignments within the iCBT program. The Beck Depression Inventory-II (BDI-II) and Kessler’s Psychological Distress Scale (K6) measures were obtained at baseline and 3-, 6-, and 12-month follow-ups. Two subgroups were defined by the median of time preference score at baseline.
Only few (835/20,000, 4.2%) workers completed the baseline survey. Of the 835 participants, 706 who fulfilled the inclusion criteria were randomly allocated to the intervention or control group. Participants who selected irrational time preference options were excluded (21 and 18 participants in the intervention and control groups, respectively). A three-way interaction (group [intervention/control] × time [baseline/follow-up] × time preference [higher/lower]) effect of iCBT was significant for BDI-II (
The effects of the iCBT were greater for the group with higher time preference at the shorter follow-up, but it was leveled off later. Workers with higher time preference may change their cognition or behavior more quickly, but these changes may not persist.
UMIN Clinical Trials Registry UMIN000014146; https://upload.umin.ac.jp/cgi-open-bin/ctr_e/ctr_view.cgi? recptno=R000016466 (Archived by WebCite at http://www.webcitation.org/70o2rNk2V)
Depressive disorder is one of the most prevalent psychiatric disorders, affecting around 340 million people worldwide [
One of the most effective psychological interventions for depression is cognitive behavioral therapy (CBT) [
Recently, variables that might predict treatment response to CBT for depression have been investigated. Previous studies have reported that the severity of depressive symptoms at baseline and the rate of change in depressive symptom severity within 5 treatment sessions significantly predicted treatment response to CBT [
Time preference (or time discounting) has attracted interest in the field of behavioral economics and behavioral medicine as a potentially common factor of multiple behaviors that pose risks for health [
This RCT aimed to examine whether an iCBT program was effective in improving nonclinical depressive symptoms among healthy workers in Japan, at 3-, 6-, and 12-month follow-ups, particularly to test the difference of the intervention effect between two subgroups classified post-hoc on the basis of time preference: a lower time preference subgroup and a higher time preference subgroup.
This study was a randomized controlled trial. The allocation ratio of the intervention group to the control group was 1:1. The Research Ethics Review Board of the Graduate School of Medicine and Faculty of Medicine, the University of Tokyo approved the study procedures (no. 3083-2). The study protocol was registered at the University Hospital Medical Information Network Clinical Trials Registry (UMIN000014146). The protocol article for this trial is available [
All workers in one corporate group (the total employee population, approximately 20,000) were recruited from one of the major telecom carrier companies in Japan by an invitation email from their internal employee assistance program staff in March 2015. Those who were interested in participating in the study were asked to go to a research website to obtain a full explanation of the study’s aim. Consent from a respondent was obtained when he or she completed a baseline questionnaire. Before the Web-based baseline survey, participants were invited to read the explanation on the research website and asked to click on an “agree” button to show their consent to participate in the study; then they proceeded to the baseline questionnaire page. Written consent was not required by the National Ethical Guidelines for Epidemiologic Research, Japan; the Research Ethics Review Board of Graduate School of Medicine and Faculty of Medicine, the University of Tokyo, approved this procedure for obtaining participants’ consents.
The inclusion criteria at the baseline survey were as follows: (1) age 20-60 years at the study entry, (2) currently employed full-time by the company, and (3) being able to access the internet via a PC at home or at their workplace. The exclusion criteria were as follows: (1) nonregular or part-time employees, (2) having an MDE in the past month, based on the diagnostic criteria on the web version of World Health Organization Composite International Diagnostic Interview 3.0 [
Participants assigned to an intervention group participated in the iCBT program called
Participants in the intervention group completed 6 weekly lessons and homework within the iCBT program. They were allowed to complete the 6 lessons and submit their homework within 10 weeks after the baseline survey. The participants were reminded by email to complete each lesson and to submit their homework if they had not already done so. Reminders were sent from the research office to the participants every Monday.
Participants in the control group were able to use an internal employee assistance program service, such as consulting with a physician or a psychologist, and group or Web-based education/training programs for promoting mental health as a treatment as usual. These programs contained few descriptions of CBT knowledge and skills.
All outcomes were measured using a Web-based self-report questionnaire at baseline and 3-, 6-, and 12-month follow-ups.
The Beck Depression Inventory-II (BDI-II) is a 21-item self-report inventory that measures depressive symptoms such as sadness, pessimism, suicidal thoughts or wishes, tiredness or fatigue, loss of energy, and loss of pleasure, among others [
Kessler’s Psychological Distress Scale (K6) consists of 6 items assessing the frequency with which respondents experienced symptoms of psychological distress during the past 30 days [
In this study, time preference was assessed by the following procedure [
In this study, we defined two subgroups according to the median time preference score at baseline because the concept of time preference has no clear cutoff point. One was the
Demographic data such as age, gender, marital status, occupation, education, and chronic disease were also collected.
We determined that to detect an effect size, a minimum sample size of 4136 in each group was necessary. This calculation considered an incidence ratio of 0.62 or greater for the onset of an MDE, at an alpha error rate of 0.05 (two-tailed) and a beta error rate of 0.20, with an expected dropout rate of 25%.
No previous study reported an effect size for a difference of intervention effect between lower and higher time preference groups. The estimated post-hoc power (1-beta) was 0.54 if the effect size was 0.2, assuming that the alpha was less than 0.05 (two-tailed), and 70% (314/448) of the initial 448 respondents in the lower time preference subgroup and 219 respondents in the higher time preference subgroup respondents completed the follow-up using the G*Power 3 program [
Participants who fulfilled the inclusion criteria were randomly assigned to an intervention or control group. Stratified permuted-block randomization was conducted as well. Participants were stratified into two strata according to K6 score (5 or greater or less than 5) on the baseline survey. A stratified permuted-block random table was generated by an independent biostatistician. Enrollment was conducted by a clinical research coordinator, and assignment was conducted by an independent research assistant. The stratified permuted-block random table was password protected and kept blind to the researcher. Only the research assistant was able to access it for random allocation. A prestratification for randomization by time preference was not conducted.
Primary analyses were conducted for the whole sample. For main analysis, a mixed model for repeated measures conditional growth model analysis was conducted to estimate the fixed effect of a three-way interaction as an indicator of intervention effect: group (intervention and control) × time (baseline and 1-, 6-, and 12-month follow-ups) × subgroup (lower time preference and higher time preference). For sensitivity analysis, a mixed model for repeated measures analysis of variance was conducted to estimate the fixed effect of three-way interaction as an indicator of intervention effect at each follow-up: group (intervention and control) × time (baseline and 1-, 6-, or 12-month follow-up) × subgroup (lower time preference and higher time preference). In these analyses, two models were applied. Model 1 was crude (not adjusted). Model 2 was adjusted by the potential confounders: gender, education, and occupation. All analyses were conducted according to the intention-to-treat principles. The MIXED procedure in SPSS Statistics 21.0 (IBM Corp, Armonk, NY, USA) was used.
Secondary analyses were conducted for all respondents as well as separately for each subgroup. A mixed model for repeated measures conditional growth model analysis was conducted to estimate the fixed effect of a group (intervention and control) × time (baseline, 1-, 6-, and 12-month follow-ups) interaction as an indicator of intervention effect. As a sensitivity analysis, a mixed model for repeated measures analysis of variance was conducted to estimate the fixed effect of a group (intervention and control) × time (baseline and 1-, 6-, or 12-month follow-up) interaction as an indicator of the intervention effect at each follow-up.
In addition, the effect sizes were calculated using estimated means based on the MIXED procedure among all respondents in each subgroup. First, estimated mean differences between baseline and follow-ups of each intervention and control group were calculated. Next, the effect sizes (ESs) were calculated by dividing between differences of the intervention and control groups by pooled SDs, which were calculated using respondents who completed the questionnaire at baseline and at follow-ups. The values of 0.2, 0.5, and 0.8 were interpreted as small, medium, and large ESs, respectively [
As a process evaluation, the rate (percentage) of completers of lessons and submitters of iCBT program homework were calculated among participants in the intervention group, for each lower and higher time preference subgroup.
Recruitment and the baseline survey were conducted in March 2015. The intervention and control groups were assessed at approximately 3 months (June 2015), 6 months (September 2015), and 12 months (March 2016) after the baseline survey.
The participant flowchart is shown in
Demographic characteristics are presented in
After excluding participants who selected irrational options (interest rate, −5% or 0%), we divided the total sample into two groups of participants: one with low levels of time preference (6% or lower annual interest rate) and one with high levels of time preference (10% or higher annual interest rate). The details of the number of respondents in each group are shown in
Participant flowchart. MDD: major depressive disorder.
Baseline characteristics of participants in the intervention and control groups, in each of the two subgroups.
Characteristic | Lower time preferencea, mean (SD) | Higher time preferenceb, mean (SD) | |||
Intervention (n=215) | Control (n=233) | Intervention (n=117) | Control (n=102) | ||
Age (years) | 38.7 (8.1) | 39.0 (7.7) | 39.6 (9.1) | 40.3 (9.3) | |
Male | 106 (49.3) | 126 (54.1) | 75 (64.1) | 69 (67.6) | |
Female | 109 (50.7) | 107 (45.9) | 42 (35.9) | 33 (32.4) | |
Never married | 91 (42.3) | 80 (34.3) | 46 (39.3) | 30 (29.4) | |
Married | 115 (53.5) | 146 (62.7) | 67 (57.3) | 67 (65.7) | |
Divorced or bereaved | 9 (4.2) | 7 (3.0) | 4 (3.4) | 5 (4.9) | |
Manager | 42 (19.5) | 45 (19.3) | 35 (29.9) | 39 (38.2) | |
Professional | 49 (22.8) | 65 (27.9) | 25 (21.4) | 20 (19.6) | |
Clerical | 101 (47.0) | 96 (41.2) | 41 (35.0) | 37 (36.3) | |
Production | 1 (0.5) | 1 (0.4) | 0 (0.0) | 0 (0.0) | |
Sales | 18 (8.4) | 20 (8.6) | 12 (10.3) | 5 (4.9) | |
Others | 4 (1.9) | 6 (2.6) | 4 (3.4) | 1 (1.0) | |
High school | 9 (4.2) | 13 (5.6) | 5 (4.3) | 5 (4.9) | |
Some college | 38 (17.7) | 38 (16.3) | 23 (19.7) | 14 (13.7) | |
University | 151 (70.2) | 165 (70.8) | 72 (61.5) | 65 (63.7) | |
Graduate school | 17 (7.9) | 17 (7.3) | 17 (14.5) | 18 (17.6) | |
Yes | 19 (8.8) | 25 (10.7) | 19 (16.2) | 16 (15.7) | |
No | 196 (91.2) | 208 (89.3) | 98 (83.8) | 86 (84.3) |
a0.1%-6% annual percentage yield.
b≥10% annual percentage yield.
In the whole sample (n=667), the iCBT program showed a significant pooled intervention effect on BDI-II (
In the lower time preference subgroup (n=448), iCBT program showed a significant pooled effect on BDI-II (
In the higher time preference subgroup (n=219), the pooled effects were not significant for both BDI-II (
Average scores of depressive symptoms (Beck Depression Inventory-II [BDI-II] and Kessler’s Psychological Distress Scale 6 [K6]) at baseline and 1-, 6-, and 12-month follow-up.
Subgroup and follow-up | Intervention | Control | ||||||||||
n | K6, mean (SD) | BDI-II, mean (SD) | n | K6, mean (SD) | BDI-II, mean (SD) | |||||||
T1b | 215 | 6.3 (4.5) | 12.7 (8.6) | 233 | 6.0 (4.3) | 12.2 (8.7) | ||||||
T2c | 143 | 5.9 (4.8) | 10.5 (8.6) | 202 | 6.0 (4.3) | 11.7 (8.2) | ||||||
T3d | 152 | 5.7 (4.6) | 10.1 (8.2) | 204 | 6.3 (5.0) | 11.6 (9.0) | ||||||
T4e | 135 | 5.6 (5.0) | 10.6 (9.5) | 187 | 6.7 (4.8) | 12.7 (9.9) | ||||||
T1 | 117 | 5.9 (5.2) | 12.2 (8.4) | 102 | 6.5 (4.9) | 13.7 (10.2) | ||||||
T2 | 78 | 4.8 (4.5) | 9.5 (7.9) | 86 | 6.7 (5.1) | 13.9 (10.9) | ||||||
T3 | 79 | 4.9 (4.5) | 9.4 (7.6) | 93 | 6.9 (5.5) | 13.9 (11.1) | ||||||
T4 | 75 | 5.9 (5.0) | 10.8 (9.9) | 86 | 6.3 (4.6) | 12.5 (10.5) |
a0.1%-6% annual percentage yield.
bT1: baseline.
cT2: 3-month follow-up.
dT3: 6-month follow-up.
eT4: 12-month follow-up.
f≥10% annual percentage yield.
Estimated mean differencea, pooled SDb, and effect sizec between groups.
Subgroup and follow-up | K6d | BDI-IIe | ||||||
Estimated Δa | Pooled SD | Effect size | Estimated Δa | Pooled SD | Effect size | |||
T2g-T1h | −0.02 | 3.69 | −0.01 | −0.98 | 6.69 | −0.15 | ||
T3i-T1 | −0.78 | 4.16 | −0.19 | −1.68 | 7.21 | −0.23 | ||
T4j-T1 | −1.21 | 4.01 | −0.30 | −2.63 | 8.02 | −0.33 | ||
T2-T1 | −1.13 | 3.77 | −0.30 | −2.55 | 6.85 | −0.37 | ||
T3-T1 | −0.88 | 4.38 | −0.20 | −2.59 | 6.36 | −0.41 | ||
T4-T1 | −0.17 | 4.40 | −0.04 | −0.91 | 7.40 | −0.12 |
aEstimated means were calculated using a MIXED procedure.
bPooled SDs were calculated using respondents those who completed the questionnaire at baseline and at follow-ups.
cEffect sizes were calculated by dividing estimated mean difference by pooled SD.
dK6: Kessler’s Psychological Distress Scale.
eBDI-II: Beck Depression Inventory-II.
f0.1%-6% annual percentage yield.
gT2, 3-month follow-up.
hT1, baseline.
iT3, 6-month follow-up.
jT4, 12-month follow-up.
k≥10% annual percentage yield.
Three-way interaction effects of the internet-based cognitive behavioral therapy, time, and time preference on Beck Depression Inventory-II (BDI-II) and Kessler’s Psychological Distress Scale (K6).
Scale and follow-up | Crude | Gender, occupational status, and education adjusted | |||||||||
Effect | 95% CI | SE | Effect | 95% CI | SE | ||||||
3 monthsa | 1.96 | 0.31 to 3.61 | 0.84 | 2.33 | .02 | 2.12 | 0.46 to 3.77 | 0.84 | 2.51 | .01 | |
6 monthsa | 0.98 | −0.67 to 2.63 | 0.84 | 1.16 | .25 | 1.14 | −0.51 to 2.79 | 0.84 | 1.35 | .18 | |
12 monthsa | −0.19 | −1.88 to 1.50 | 0.86 | −0.22 | .83 | −0.01 | −1.70 to 1.68 | 0.86 | −0.01 | .99 | |
Pooledb | −0.06 | −0.45 to 0.34 | 0.20 | −0.27 | .79 | −0.02 | −0.42 to 0.38 | 0.20 | −0.11 | .91 | |
3 monthsa | 3.57 | 0.43 to 6.72 | 1.60 | 2.23 | .03 | 3.75 | 0.60 to 6.91 | 1.61 | 2.33 | .02 | |
6 monthsa | 2.87 | −0.28 to 6.01 | 1.60 | 1.79 | .07 | 3.07 | −0.08 to 6.22 | 1.61 | 1.91 | .06 | |
12 monthsa | 0.23 | −2.97 to 3.44 | 1.64 | 0.14 | .89 | 0.44 | −2.78 to 3.65 | 1.64 | 0.27 | .79 | |
Pooledb | 0.01 | −0.75 to 0.78 | 0.39 | 0.04 | .97 | 0.05 | −0.72 to 0.82 | 0.39 | 0.13 | .90 |
aA mixed model for repeated measures analysis of variance model analyses was conducted to estimate a three-way interaction effect among intervention, time, and time preference.
bA mixed model for repeated measures conditional growth model analyses was conducted to estimate a three-way interaction effect.
Progress of learning in the internet-based cognitive behavioral therapy program in the two subgroups.
Contents |
Lower time preference (n=215), n (%) | Higher time preference (n=117), n (%) | ||
Completers of lessons | Submitters of homework | Completers of lessons | Submitters of homework | |
Lesson (L)1: |
186 (86.5) | 124 (57.7) | 103 (88.0) | 70 (59.8) |
L2: |
181 (84.2) | 85 (39.5) | 97 (82.9) | 51 (43.6) |
L3: |
167 (77.7) | 84 (39.1) | 87 (74.4) | 48 (41.0) |
L4: |
153 (71.2) | 68 (31.6) | 80 (68.4) | 37 (31.6) |
L5: |
142 (66.0) | 54 (25.1) | 76 (65.0) | 34 (29.1) |
L6: |
138 (64.2) | 55 (25.6) | 69 (59.0) | 28 (23.9) |
All 6 lessons | 136 (63.3) | 37 (17.2) | 68 (58.1) | 20 (17.1) |
This RCT examined the effects of iCBT on improving nonclinical depressive symptoms at 3-, 6-, and 12-month follow-ups among healthy workers by lower and higher time preference subgroups in Japan. As a result, the three-way interaction effect of iCBT was significant for nonclinical depressive symptoms at 3-month follow-up, after adjusting for gender, occupational status, and education. In the higher time preference subgroup, iCBT showed a significant intervention effect on nonclinical depressive symptoms at 3- and 6-month follow-ups, while the pooled effect was not significant. On the other hand, in the lower time preference subgroup, iCBT showed significant ESs on nonclinical depressive symptoms at 6- and 12-month follow-ups. The iCBT program showed a significant pooled effect on nonclinical depressive symptoms at 12-month follow-up.
To our knowledge, this is the first RCT that has demonstrated the effect of an iCBT on improving nonclinical depressive symptoms, specifically targeting workers with lower or higher time preference. iCBT showed a significantly higher effect for improving nonclinical depressive symptoms in the higher time preference subgroup than the lower time preference subgroup at 3-month follow-up. Workers with higher time preference may more easily change their cognition or behavior, but these changes persisted for only a short period. The pooled effect of iCBT was significant only in the lower time preference subgroup. Workers with lower time preference may be more likely to keep their cognitive or behavioral changes for a longer period.
This study showed a difference in the intervention effect of iCBT between the higher time preference subgroup and the lower time preference subgroup. However, in the process evaluation, there were no differences between completers of lessons and submitters of homework of the iCBT program in both subgroups. Our findings caused us to reject the hypothesis that participants with higher time preference were less likely to follow the program.
Previous systematic reviews suggested that higher time preference was associated with poor responses to health promotion interventions such as dietary and weight loss programs [
The intervention effects of iCBT were less persistent among workers with higher long-term time preferences (eg, over 6 months). These findings support the hypothesis that the effect of CBT is not persistent among people with higher time preferences. Workers with higher time preferences may experience difficulty in maintaining their cognitive and behavioral changes. Workers with higher time preferences may stop using their new CBT-related perspectives or behaviors when their problems are solved (ie, improvement of nonclinical depressive symptoms). They may underestimate the future risk for a recurrence of the problems and not keep practicing a preventive effort. A follow-up program providing incentives (eg, allocating points or giving a prize as a reward) may reinforce continuing activities, making the iCBT program more effective even after 6 months for workers with higher time preferences.
These findings may contribute to further understanding of behavioral characteristics of people based on their (higher or lower) time preference. In this study, workers with higher time preferences were less likely to maintain the effects of a CBT-based program over the long-term, compared with those with lower time preference, while both groups engaged in learning to a similar extent. This pattern was consistent with previous reports on the impact of time preferences on health-related behaviors such as obesity and smoking [
There are several limitations of this study that should be considered. First, we did not conduct a prestratification for randomization by time preference. The sample may be biased between the intervention and control groups in each subgroup. Second, participants were recruited from one corporate group in Japan. The participation rate was very low (835/20,000, 4.2%). Most participants were married, working in clerical positions, and university graduates. They had their own PCs or tablet computers in their offices or homes. The participants were also supposed to have experience using a PC and studying through Web-based programs. Higher education level may also help participants learn from the iCBT program. The generalization of these findings to the general working population is limited. Third, while we excluded those who had MDE before, the scores of depressive symptoms and psychological distress of the participants at baseline were relatively high. These findings may be more applicable to respondents with mild depression. Fourth, the dropout rates in this study were 27.9% (197/706), 25.2% (178/706), and 31.6% (223/706) at the 3-, 6-, and 12-month follow-ups, respectively. The dropout rates were higher in the intervention group than in the control group during the entire follow-up period. The dropouts may have caused a selection bias, particularly if the intervention group participants with higher levels of depression were more likely to quit the program. Fifth, it is possible that participants in the control group acquired information about the iCBT program from participants in the intervention group at the same workplace. This contamination could weaken the intervention effect. Sixth, all outcomes in this study were measured by self-report, which might have been affected by the perception of participants or by situational factors at work.
The iCBT program was significantly better at improving nonclinical depressive symptoms in the higher time preference subgroup compared with the lower time preference subgroup at the 3-month follow-up. Workers with higher time preferences may easily change their cognition or behavior, but the change may persist for only a short period. On the other hand, the pooled effect of iCBT during the entire follow-up period was significant only in the lower time preference subgroup. Workers with lower time preferences may be likely to keep their cognitive or behavioral changes for a longer period. A further RCT with a precise design, such as stratified permuted-block randomization, should be conducted to test the potential different intervention effects of the iCBT program on nonclinical depressive symptoms between lower and higher time preference subgroups.
The time preference questionnaire used in the study and the number of respondents in each category.
CONSORT-EHEALTH checklist (V 1.6.1).
Beck Depression Inventory-II
cognitive behavioral therapy
effect size
internet-based computerized cognitive behavioral therapy
Kessler’s Psychological Distress Scale
major depressive disorder
major depressive episode
randomized controlled trial
The authors appreciate the help of the following persons in completing this project: Takayuki Narumi, Jun Naoi, Keisuke Kito, Chinatsu Narumi, Hayato Mori, Mitsuyasu Mizusaki, Chihiro Hoshino, and Aya Matsumoto.
KI and NK conceived of the study, developed study design, conducted literature search, collected, analyzed, and interpreted data, and prepared the first draft. TAF, YM, AS, KKuribayashi, and KKasai developed study design and interpreted data. All authors reviewed the manuscript.
TAF has received lecture fees from Janssen, Meiji, Mitsubishi-Tanabe, MSD KK, and Pfizer. He has received research support from Mitsubishi-Tanabe. NK reports grants from Infocom Corp, Fujitsu Ltd, Fujitsu Software Technologies, and TAK Ltd, personal fees from Occupational Health Foundation, Japan Dental Association, Sekisui Chemicals, Junpukai Health Care Center, Osaka Chamber of Commerce and Industry, outside the submitted work. The other authors declare that they have no competing interests.