Abstract
Risk prediction models offer a promising approach to lifestyle modification. We evaluated the effect of personalized advice based on cancer risk prediction in improving five lifestyle habits (smoking, alcohol consumption, salt intake, physical activity, and body mass index) compared with standard advice without risk prediction among a Japanese general population with at least one unhealthy lifestyle habit.
In a parallel-design, single-blind, randomized controlled trial between February 2018 and July 2019, 5984 participants aged 40–64 years with unhealthy lifestyle habits were recruited from persons covered under a life insurance policy. They were randomly assigned to an intervention or control group and received personalized or standard advice, respectively. They were also sent an invitation to participate in a lifestyle modification program aimed at improving lifestyle. Primary outcome was an improvement in lifestyle, defined as an increase in healthy lifestyle habits within 6 months.
The proportion of participants who improved their lifestyle within 6 months in the intervention group did not significantly differ from that in the control group (18.4% vs. 17.7%; P = 0.488). Among participants with low health literacy and two or fewer of five healthy habits, the proportion of participants subscribing to the lifestyle modification program was higher in the intervention group than in the control group.
Compared with standardized advice, personalized advice based on cancer risk prediction had no effect on improving lifestyle.
Provision of predicted cancer risk information did not induce change in unhealthy lifestyle.
Introduction
There were 24.5 million incident cancer cases worldwide in 2017 (1). Previous studies have estimated that up to 45% of cancer incidence is attributable to modifiable factors such as smoking, alcohol, and physical activity (2, 3). For primary prevention of cancer, although several recommendations on lifestyle have been provided (4–6), gaps exist between the recommendation and the actual behavior of individuals. For example, comprehensive, evidence-based recommendations for cancer prevention applicable for Japanese people have been provided as ‘Cancer Prevention Recommendation for Japanese’ (6), which formed the basis of the Ministry of Health, Labour and Welfare's Basic Plan to Promote Cancer Control Program (7). However, the proportion of individuals who adhered to all these healthy habits in one large population was less than 3% in men and approximately 20% in women (8).
Many risk prediction models for cancer occurrence based on lifestyle and genetic factors have been developed (9–11). The provision of information on the probability of cancer occurrence using such prediction models is a promising approach to reminding people about their risk of cancer and then motivating them to modify unhealthy lifestyle habits (12, 13). However, evidence for the effectiveness of such information on the modification of lifestyle habits has been found to be limited in general population settings (9, 14). A recent study showed that personalized cancer risk information with lifestyle advice had no effect on changing behavior among a general population recruited through the Internet (15). In a worksite setting, in contrast, health risk feedback along with a health promotion program had a meaningful effect on improvement in lifestyle habits, such as smoking, alcohol use, and dietary fat intake (16). We have speculated that advice based on personalized cancer risk in general populations might motivate individuals to modify their lifestyle and increase participation in health promotion programs and, as a result, choose a healthy lifestyle to reduce their cancer risk. An important task for precision public health is to elucidate whether such personalized intervention using risk prediction models is effective in improving lifestyle and reducing cancer incidence (17).
We previously developed a prediction model of 10-year cumulative cancer risk in improving five lifestyle habits, namely smoking habit, drinking habit, salt intake, physical activity, and body mass index (BMI; ref. 11). Here, we conducted a randomized controlled trial to evaluate the effectiveness of personalized advice based on this prediction model in comparison with standardized advice for cancer prevention not based on this prediction model in a general population with at least one unhealthy lifestyle habit.
Materials and Methods
Study design
This intervention study was a parallel-design, single-blind, randomized controlled trial in individuals with one or more unhealthy lifestyle habits (UMIN-CTR, number UMIN000031313; ref. 18). Participants were allocated to either an intervention group which was given personalized advice based on probability calculated by a prediction model of 10-year cumulative cancer risk, or a control group which was given standardized advice for the prevention of cancer.
The protocol was approved by the National Cancer Center Institutional Review Board. The study design did not deviate from the original protocol.
Participants
Between February and April, 2018, a consent form and a paper-based baseline questionnaire were sent from The Dai-ichi Life Insurance Company, Limited (Dai-ichi Life) to individuals as follows: individuals who bought a life insurance policy from Dai-ichi Life in November or December, 2017; individuals who were both the policy holder and the person insured on the policy; and individuals who were between aged 40 and 64 years on January 1, 2018. The baseline questionnaire, which was based on a previous study (19), assessed current smoking status, average frequencies and quantities of drinking alcohol, average frequencies of eating fish roe, average time per day spent in each of three activities (heavy physical work and strenuous exercise, sitting, and walking or standing), and height and weight to estimate five lifestyle habits. This questionnaire also assessed age, sex, clinical history of six diseases (cancer, heart disease, stroke, hypertension, diabetes, and chronic renal failure), health literacy, marital status, education, occupation, employment status, and household income. Questions on health literacy consisted of five items which asked whether the participants would have the ability to (i) collect health-related information from various sources, (ii) extract the information requested, (iii) understand and communicate the obtained information, (iv) consider the credibility of the information and (v) make decisions based on the information (20). Each item was rated on a 5-point scale ranging from 1, “strongly disagree”, to 5, “strongly agree”.
The consent form provided to all potential participants clearly stated that Dai-ichi Life would not subject individuals meeting the inclusion criteria to any penalty or disadvantage with regard to their current and any future insurance policy no matter whether they participated in the study or not. Although enrolled participants were insured under the company's policies, Dai-ichi Life played no further role in the procedures and communications of the study following its initial sending of the consent form and baseline questionnaire. The participants returned their completed questionnaires directly to the National Cancer Center (NCC). All other procedures and communications were between the participant and NCC.
On the basis of a previous study (8), the healthy lifestyle habits were defined as follows: not smoking was defined as never smoker or past smoker; moderate or no alcohol consumption was defined as alcohol consumption of < 150 g/week; moderate salt intake was defined as consumption of < 0.67 g of fish roe per day, which corresponded to eating fish roe less than once a week in the questionnaire; adequate physical activity was defined as ≥37.5 and ≥31.9 metabolic equivalent hours per day (MET-hour/day) for men and women, respectively; and appropriate BMI was defined as a BMI within the range 21–27 kg/m2 and 19–25 kg/m2 for men and women, respectively. The threshold of smoking habit, alcohol consumption, and BMI was directly recommended by ‘Cancer Prevention Recommendation for Japanese'. The threshold was set through a systematic review of the literature, meta-analysis of published data, and pooled analysis of cohort studies in Japan (6). The threshold of salt intake and physical activity was set in reference to previous studies: elevated cancer risk was observed above the 4th quintile group of fish roe intake (21), which corresponded to the threshold of moderate salt intake. Reduced cancer risk was observed among the highest quartile group of daily total physical activity (22), which corresponded to the threshold of adequate physical activity. The unhealthy lifestyle habits were defined as lifestyle habits, which did not adhere to the healthy ones. Individuals who provided written informed consent and their address, which was necessary to send the advice and follow-up questionnaires, and who had one or more unhealthy lifestyle habits were eligible for this study. Participants with a clinical history of cancer were not excluded on the basis that the recommended lifestyle to prevent cancer might improve survival and reduce the future risk of cancer (23).
Data sharing
The consent form specified the data sharing practices agreed between NCC and Dai-ichi Life, as follows. The questionnaire, sent from Dai-ichi Life, was assigned an individual number to allow identification of each potential participant and data sharing. NCC would inform Dai-ichi Life of (i) the answers to the questionnaire, excluding those concerning clinical history and socio-economic status; and (ii) the status of subscription to the lifestyle modification program, for use in developing new insurance products or services only. Dai-ichi Life would inform NCC of the insured person's usage history of a smartphone app-based cancer risk prediction tool to allow assessment of the effect of exposure to information on cancer risk other than the cancer risk provided for in this study.
Randomization
Eligible individuals were randomly assigned in a 1:1 ratio to the intervention or control group via computerized stratified blocked randomization with block sizes of two. The randomization was stratified by age (40–49, 50–59, or 60–64 years) and sex based on the baseline questionnaire and was implemented after consent. Use of a computer program for randomization allowed the allocation to be concealed, and eliminated selection bias. Participants were not masked because they could determine their allocated group based on the advice received. Researchers other than research staff who sent the advice were masked to the allocation.
Procedures
Participants in the control group were sent paper-based standard advice. The standard advice recommended improving their unhealthy lifestyle habits based on “Cancer Prevention Recommendation for Japanese,” which addresses healthy lifestyle habits, such as not smoking, drinking moderately, restricting salt intake, being active in daily life, and maintaining an appropriate weight (6). Participants in the intervention group were sent paper-based personalized advice. The personalized advice included (i) the probability of cancer occurrence calculated by a prediction model of 10-year cumulative cancer risk based on the pattern of five lifestyle habits (11); (ii) the most influential habit, namely the lifestyle habit having the greatest influence on the probability of cancer occurrence, based on the prediction model; and (iii) the probability of cancer occurrence if the most influential lifestyle habit was improved, in addition to the standard advice received by the control group. These probabilities were presented as a bar chart to facilitate understanding of the effect of improving the most influential unhealthy lifestyle habit. Although the prediction model was provided through the NCC's website or Dai-ichi Life's smartphone app independently of the study, participants were not proactively informed that it was available.
Along with the advice, participants of both groups were sent an invitation to participate in a lifestyle modification program aimed at improving their unhealthy lifestyle habits which used paper-based materials. This program was cost-free and intended to remind about the benefits, reduce barriers, increase self-efficacy, and become a cue to improving unhealthy lifestyle habits. Participants could subscribe to any one of five courses – quitting smoking, reducing alcohol consumption, improving diet, being more active, and losing weight – related to their unhealthy lifestyle habits within 6 months after receiving the advice. Participants in the intervention group were proposed a recommended course that related to the most influential unhealthy lifestyle habit in the personalized advice. However, they were free to choose a different course if they preferred. In this program, subscribers were sent paper-based materials that provided information on the risks of unhealthy lifestyle habits for cancer occurrence, practical methods to improve unhealthy lifestyle habits and maintain healthy ones, and health literacy skills. The materials were sent to subscribers once a month between subscription to the course and 12 months after receiving the advice. In addition, subscribers were able to access consultation on their health-related problems by telephone.
The 6-month and 12-month questionnaires were conducted as follow-up questionnaires for participants at 6 months and 12 months after receiving the advice, respectively. The follow-up questionnaires assessed the five lifestyle habits and clinical history after the advice had been sent. Participants could select paper-based or web-based questionnaires in the follow-up questionnaires. After the 12-month questionnaire, personalized advice based on responses to the 12-month questionnaire was sent to both groups.
Outcomes
The primary outcome was improvement in lifestyle, which was defined as an increase in the number of healthy lifestyle habits occurring within 6 months after sending the advice. Secondary outcomes were improvement in lifestyle within 12 months after sending the advice, subscription to the lifestyle modification program, improvement in each lifestyle habit within 6 and 12 months after sending the advice, and change in continuous variables, namely alcohol consumption (g/week), physical activity (MET-hour/day), and BMI (kg/m2), within 6 and 12 months after sending the advice.
Statistical analysis
On the basis of a previous study assessing the effectiveness of personalized risk information on attendance at a stop smoking service in England (24), which was similar to the design of our study, we assumed that the proportion of participants who would subscribe to the life modification program would be 15% in the intervention group and 10% in the control group. Furthermore, we assumed that one-third of subscribers to the life modification program would improve their unhealthy lifestyle habits based on a previous study (25). On the basis of a systematic review which reported that the effect magnitude of a change in unhealthy lifestyle habits by provision of health risk information with feedback only was small (16), we assumed that no participants would improve their unhealthy lifestyle habits if they did not subscribe to the lifestyle modification program. From this, we calculated that the proportion of participants who would improve their unhealthy lifestyle habits would be 5% in the intervention group and 3.3% in the control group. This in turn indicated that 3,000 participants per group with a 20% dropout rate for the 6-month questionnaire were required to detect a 1.7% difference at a 5% significance level (two-sided) with 80% power. In a previous study in Japan, the proportion of individuals who adhered to all five healthy lifestyle habits was approximately 12% (8). Thus, we planned to collect 7,000 individuals who returned the baseline questionnaire in the study.
Evaluations were performed on the intention-to-treat principal, except for the analyses of changes in continuous variables, in which participants who did not answer the baseline questions were excluded from the intention-to-treat population. When participants did not respond to questions on the lifestyle habits necessary to classify them as healthy or not at baseline, they were classified as healthy. When participants did not respond to questions at follow-up, the answers given in the previous questionnaire were imputed. Characteristics at baseline of the intervention and control groups were presented using descriptive statistics. Health literacy was reported as the mean of scales on the five items.
For the primary outcome, the proportion of participants who improved their lifestyle within 6 months in the intervention group was compared to that in the control group using the Cochran–Mantel–Haenszel test stratified by age and sex. The odds ratio was adjusted for age and sex by the Mantel–Haenszel method. Multivariable-adjusted ORs were estimated using a logistic regression model to adjust for some factors at baseline, namely age, sex, number of healthy lifestyle habits, clinical history, health literacy, marital status, education, occupation, employment status, household income, and usage history of a cancer risk prediction tool, which was a component of a smartphone app. Subscription to the lifestyle modification program was not included in the adjustment because the subscription was influenced by the intervention. We conducted a sensitivity analysis among participants who answered the 6-month questionnaire to assess bias from dropout at 6 months on primary outcome.
For the secondary outcomes, the proportion of participants in the intervention group who improved their lifestyle within 12 months, subscribed to the lifestyle modification program, and improved each lifestyle habit was compared to that in the control group by the χ2 test. Multivariable-adjusted ORs were estimated in the same way as for the primary outcome. In addition, the difference between the intervention and control groups in change in the continuous variables of alcohol consumption, physical activity, and BMI within 6 and 12 months was estimated and compared by the t test. Multivariable-adjusted differences of change in the continuous variables were estimated using a linear regression model to adjust for the same factors as in the primary outcome.
For prespecified subgroup analyses, intervention effects on the primary and secondary outcomes were estimated in the different groups of age, sex, number of healthy lifestyle habits, clinical history, health literacy, marital status, education, occupation, employment status, household income, and usage history of the cancer risk prediction tool in a smartphone app at baseline. For additional analyses, the proportion of subscribers of the lifestyle modification program who chose the course related to the most influential lifestyle habit for preventing cancer were compared descriptively.
Analyses were conducted using SAS version 9.4 (SAS Institute Inc.).
Results
Between February, 2018 and July, 2019, 6534 participants were enrolled in the study, of whom 5,894 with one or more unhealthy lifestyle habits were randomly assigned to the intervention or control group (n = 2947 each) (Fig. 1). Except for three participants who provided the wrong address in the consent form, the advice was sent to 5891 participants, who comprised the intention-to-treat population. Of these, 4,678 (79.4%) participants and 4,583 (77.8%) participants completed the 6- and 12-month questionnaires, respectively.
Table 1 shows baseline characteristics. Mean age was 51.3 years (SD, 6.7) in the intervention group and 51.2 years (6.6) in the control group. A total of 1,565 (53.1%) participants in the intervention group and 1563 (53.1%) in the control group were male.
. | Intervention group (n = 2,947) . | Control group (n = 2,944) . |
---|---|---|
Age, mean (SD), y | 51.3 (6.7) | 51.2 (6.6) |
40–49 y | 1,326 (45.0%) | 1,326 (45.0%) |
50–59 y | 1,259 (42.7%) | 1,257 (42.7%) |
60–64 y | 362 (12.3%) | 361 (12.3%) |
Sex | ||
Male | 1,565 (53.1%) | 1,563 (53.1%) |
Female | 1,382 (46.9%) | 1,381 (46.9%) |
Alcohol consumption, median (IQR), g/week | 35 (0–221) | 35 (0–231) |
Physical activity, mean (SD), MET-hour/day | 32.1 (6.1) | 32.2 (6.2) |
BMI, mean (SD), kg/m2 | 23.2 (3.7) | 23.2 (3.7) |
Healthy lifestyle | ||
No smoking | 2,276 (77.2%) | 2,248 (76.4%) |
Moderate or no alcohol consumption | 1,949 (66.1%) | 1,941 (65.9%) |
Moderate salt intake | 2,698 (91.6%) | 2,710 (92.1%) |
Adequate physical activity | 896 (30.4%) | 914 (31.0%) |
Appropriate BMI | 1,461 (49.6%) | 1,447 (49.2%) |
Number of healthy lifestyle habits | ||
0 | 7 (0.2%) | 6 (0.2%) |
1 | 109 (3.7%) | 104 (3.5%) |
2 | 481 (16.3%) | 500 (17.0%) |
3 | 1,191 (40.4%) | 1,180 (40.1%) |
4 | 1,159 (39.3%) | 1,154 (39.2%) |
Clinical history | ||
None | 2,280 (77.4%) | 2,279 (77.4%) |
Any | 667 (22.6%) | 665 (22.6%) |
Cancer | 56 (1.9%) | 48 (1.6%) |
Heart disease | 52 (1.8%) | 39 (1.3%) |
Stroke | 21 (0.7%) | 34 (1.2%) |
Hypertension | 501 (17.0%) | 524 (17.8%) |
Diabetes | 132 (4.5%) | 121 (4.1%) |
Chronic renal failure | 5 (0.2%) | 5 (0.2%) |
Health literacy, mean (SD) | 3.5 (0.7) | 3.5 (0.7) |
Marital status | ||
With a spouse | 2,119 (71.9%) | 2,103 (71.4%) |
Without a spouse | 818 (27.8%) | 826 (28.1%) |
Education | ||
High school or less | 1,300 (44.1%) | 1,312 (44.6%) |
College | 825 (28.0%) | 805 (27.3%) |
University or higher | 816 (27.7%) | 815 (27.7%) |
Occupation | ||
Unemployed or homemaker | 255 (8.7%) | 285 (9.7%) |
White collar | 2,127 (72.2%) | 2,130 (72.4%) |
Blue collar | 447 (15.2%) | 432 (14.7%) |
Employment status | ||
Permanent | 1,744 (59.2%) | 1,736 (59.0%) |
Fixed-term | 707 (24.0%) | 651 (22.1%) |
Self-employed or administrator | 280 (9.5%) | 306 (10.4%) |
House income | ||
<3 million yen | 532 (18.1%) | 511 (17.4%) |
3–6 million yen | 1,157 (39.3%) | 1,086 (36.9%) |
6–9 million yen | 751 (25.5%) | 792 (26.9%) |
≥9 million yen | 486 (16.5%) | 532 (18.1%) |
Usage history of a smartphone app | ||
Yes | 243 (8.2%) | 199 (6.8%) |
No | 2,704 (91.8%) | 2,745 (93.2%) |
. | Intervention group (n = 2,947) . | Control group (n = 2,944) . |
---|---|---|
Age, mean (SD), y | 51.3 (6.7) | 51.2 (6.6) |
40–49 y | 1,326 (45.0%) | 1,326 (45.0%) |
50–59 y | 1,259 (42.7%) | 1,257 (42.7%) |
60–64 y | 362 (12.3%) | 361 (12.3%) |
Sex | ||
Male | 1,565 (53.1%) | 1,563 (53.1%) |
Female | 1,382 (46.9%) | 1,381 (46.9%) |
Alcohol consumption, median (IQR), g/week | 35 (0–221) | 35 (0–231) |
Physical activity, mean (SD), MET-hour/day | 32.1 (6.1) | 32.2 (6.2) |
BMI, mean (SD), kg/m2 | 23.2 (3.7) | 23.2 (3.7) |
Healthy lifestyle | ||
No smoking | 2,276 (77.2%) | 2,248 (76.4%) |
Moderate or no alcohol consumption | 1,949 (66.1%) | 1,941 (65.9%) |
Moderate salt intake | 2,698 (91.6%) | 2,710 (92.1%) |
Adequate physical activity | 896 (30.4%) | 914 (31.0%) |
Appropriate BMI | 1,461 (49.6%) | 1,447 (49.2%) |
Number of healthy lifestyle habits | ||
0 | 7 (0.2%) | 6 (0.2%) |
1 | 109 (3.7%) | 104 (3.5%) |
2 | 481 (16.3%) | 500 (17.0%) |
3 | 1,191 (40.4%) | 1,180 (40.1%) |
4 | 1,159 (39.3%) | 1,154 (39.2%) |
Clinical history | ||
None | 2,280 (77.4%) | 2,279 (77.4%) |
Any | 667 (22.6%) | 665 (22.6%) |
Cancer | 56 (1.9%) | 48 (1.6%) |
Heart disease | 52 (1.8%) | 39 (1.3%) |
Stroke | 21 (0.7%) | 34 (1.2%) |
Hypertension | 501 (17.0%) | 524 (17.8%) |
Diabetes | 132 (4.5%) | 121 (4.1%) |
Chronic renal failure | 5 (0.2%) | 5 (0.2%) |
Health literacy, mean (SD) | 3.5 (0.7) | 3.5 (0.7) |
Marital status | ||
With a spouse | 2,119 (71.9%) | 2,103 (71.4%) |
Without a spouse | 818 (27.8%) | 826 (28.1%) |
Education | ||
High school or less | 1,300 (44.1%) | 1,312 (44.6%) |
College | 825 (28.0%) | 805 (27.3%) |
University or higher | 816 (27.7%) | 815 (27.7%) |
Occupation | ||
Unemployed or homemaker | 255 (8.7%) | 285 (9.7%) |
White collar | 2,127 (72.2%) | 2,130 (72.4%) |
Blue collar | 447 (15.2%) | 432 (14.7%) |
Employment status | ||
Permanent | 1,744 (59.2%) | 1,736 (59.0%) |
Fixed-term | 707 (24.0%) | 651 (22.1%) |
Self-employed or administrator | 280 (9.5%) | 306 (10.4%) |
House income | ||
<3 million yen | 532 (18.1%) | 511 (17.4%) |
3–6 million yen | 1,157 (39.3%) | 1,086 (36.9%) |
6–9 million yen | 751 (25.5%) | 792 (26.9%) |
≥9 million yen | 486 (16.5%) | 532 (18.1%) |
Usage history of a smartphone app | ||
Yes | 243 (8.2%) | 199 (6.8%) |
No | 2,704 (91.8%) | 2,745 (93.2%) |
Note: Data are number (%), unless otherwise specified.
Abbreviations: SD, standard deviation; IQR, interquartile range; MET, metabolic equivalent; BMI, body mass index.
The primary outcome in the intervention group was not significantly different from that in the control group [proportion of participants who improved their lifestyle within 6 months, 18.4% vs. 17.7%; OR, 1.05; 95% confidence interval (CI), 0.92–1.20; P = 0.488; Table 2]. After adjusting for baseline factors, the personalized advice did not improve lifestyle compared with the standard advice (multivariable-adjusted OR, 1.01; 95% CI, 0.87–1.17; P = 0.920). In the sensitivity analysis, there was no significant difference between groups in the proportion of participants answering the 6-month questionnaire (23.5% vs. 22.3%; OR, 1.08; 95% CI, 0.94–1.23; P = 0.291). In the secondary outcome analyses, there were no statistically significant differences between the groups (Tables 2 and 3), except for change in physical activity between baseline and 12 months, which was significantly higher in the intervention than in the control group, but nevertheless not a meaningful effect of the difference between the two groups in the proportion of participants who improved their physical activity (difference of change, 0.29 MET-hour/day; 95% CI, 0.03–0.55 MET-hour/day; P = 0.027).
. | Intervention group (n = 2,947) . | Control group (n = 2,944) . | OR (95% CI) . | P . | Adjusted OR (95% CI) . | P . |
---|---|---|---|---|---|---|
Improvement in lifestyle within 6 months | 543 (18.4%) | 522 (17.7%) | 1.05 (0.92–1.20) | 0.488 | 1.01 (0.87–1.17) | 0.920 |
Improvement in lifestyle within 12 months | 610 (20.7%) | 562 (19.1%) | 1.11 (0.97–1.26) | 0.122 | 1.10 (0.95–1.27) | 0.197 |
Subscription to the lifestyle modification program | 784 (26.6%) | 727 (24.7%) | 1.11 (0.98–1.24) | 0.093 | 1.07 (0.94–1.21) | 0.323 |
Quitting smoking | 74 | 31 | ||||
Reducing alcohol consumption | 109 | 67 | ||||
Improving diet | 127 | 143 | ||||
Being more active | 402 | 320 | ||||
Losing weight | 72 | 166 | ||||
Improvement in each lifestyle habit within 6 months | ||||||
Smoking | 26 (0.9%) | 33 (1.1%) | 0.79 (0.47–1.32) | 0.358 | 0.74 (0.42–1.30) | 0.298 |
Drinking | 130 (4.4%) | 109 (3.7%) | 1.20 (0.93–1.56) | 0.168 | 1.21 (0.92–1.59) | 0.175 |
Salt intake | 118 (4.0%) | 93 (3.2%) | 1.28 (0.97–1.69) | 0.081 | 1.17 (0.86–1.57) | 0.315 |
Physical activity | 298 (10.1%) | 303 (10.3%) | 0.98 (0.83–1.16) | 0.819 | 0.96 (0.79–1.16) | 0.674 |
BMI | 132 (4.5%) | 135 (4.6%) | 0.98 (0.76–1.25) | 0.844 | 0.90 (0.69–1.18) | 0.461 |
Improvement in each lifestyle habit within 12 months | ||||||
Smoking | 41 (1.4%) | 43 (1.5%) | 0.95 (0.62–1.46) | 0.822 | 0.93 (0.58–1.50) | 0.768 |
Drinking | 144 (4.9%) | 124 (4.2%) | 1.17 (0.91–1.49) | 0.214 | 1.16 (0.89–1.51) | 0.270 |
Salt intake | 134 (4.5%) | 124 (4.2%) | 1.08 (0.84–1.39) | 0.530 | 1.00 (0.76–1.31) | 0.994 |
Physical activity | 319 (10.8%) | 291 (9.9%) | 1.11 (0.94–1.31) | 0.236 | 1.08 (0.90–1.31) | 0.402 |
BMI | 164 (5.6%) | 172 (5.8%) | 0.95 (0.76–1.18) | 0.646 | 0.96 (0.76–1.22) | 0.756 |
. | Intervention group (n = 2,947) . | Control group (n = 2,944) . | OR (95% CI) . | P . | Adjusted OR (95% CI) . | P . |
---|---|---|---|---|---|---|
Improvement in lifestyle within 6 months | 543 (18.4%) | 522 (17.7%) | 1.05 (0.92–1.20) | 0.488 | 1.01 (0.87–1.17) | 0.920 |
Improvement in lifestyle within 12 months | 610 (20.7%) | 562 (19.1%) | 1.11 (0.97–1.26) | 0.122 | 1.10 (0.95–1.27) | 0.197 |
Subscription to the lifestyle modification program | 784 (26.6%) | 727 (24.7%) | 1.11 (0.98–1.24) | 0.093 | 1.07 (0.94–1.21) | 0.323 |
Quitting smoking | 74 | 31 | ||||
Reducing alcohol consumption | 109 | 67 | ||||
Improving diet | 127 | 143 | ||||
Being more active | 402 | 320 | ||||
Losing weight | 72 | 166 | ||||
Improvement in each lifestyle habit within 6 months | ||||||
Smoking | 26 (0.9%) | 33 (1.1%) | 0.79 (0.47–1.32) | 0.358 | 0.74 (0.42–1.30) | 0.298 |
Drinking | 130 (4.4%) | 109 (3.7%) | 1.20 (0.93–1.56) | 0.168 | 1.21 (0.92–1.59) | 0.175 |
Salt intake | 118 (4.0%) | 93 (3.2%) | 1.28 (0.97–1.69) | 0.081 | 1.17 (0.86–1.57) | 0.315 |
Physical activity | 298 (10.1%) | 303 (10.3%) | 0.98 (0.83–1.16) | 0.819 | 0.96 (0.79–1.16) | 0.674 |
BMI | 132 (4.5%) | 135 (4.6%) | 0.98 (0.76–1.25) | 0.844 | 0.90 (0.69–1.18) | 0.461 |
Improvement in each lifestyle habit within 12 months | ||||||
Smoking | 41 (1.4%) | 43 (1.5%) | 0.95 (0.62–1.46) | 0.822 | 0.93 (0.58–1.50) | 0.768 |
Drinking | 144 (4.9%) | 124 (4.2%) | 1.17 (0.91–1.49) | 0.214 | 1.16 (0.89–1.51) | 0.270 |
Salt intake | 134 (4.5%) | 124 (4.2%) | 1.08 (0.84–1.39) | 0.530 | 1.00 (0.76–1.31) | 0.994 |
Physical activity | 319 (10.8%) | 291 (9.9%) | 1.11 (0.94–1.31) | 0.236 | 1.08 (0.90–1.31) | 0.402 |
BMI | 164 (5.6%) | 172 (5.8%) | 0.95 (0.76–1.18) | 0.646 | 0.96 (0.76–1.22) | 0.756 |
Note: Data are number (%), unless otherwise specified. OR on improvement in lifestyle within 6 months is adjusted for age and sex by the Mantel–Haenszel method. Adjusted OR is adjusted for age, sex, number of healthy lifestyle habits, clinical history, health literacy, marital status, education, occupation, employment status, household income, and usage history of the cancer risk prediction tool in a smartphone app.
Abbreviations: BMI, body mass index; OR, odds ratio; CI, confidence interval.
. | Intervention group (n = 2,947) . | Control group (n = 2,944) . | Difference in change (Intervention group − Control group) . | P . | Adjusted difference of change (Intervention group − Control group) . | P . |
---|---|---|---|---|---|---|
Change within 6 months | ||||||
Alcohol consumption (g/week) | −2.75 (−6.33–0.82) | −5.25 (−9.45 to −1.06) | 2.50 (−3.01–8.01) | 0.375 | 2.59 (−3.22–8.40) | 0.382 |
Physical activity (MET-hour/day) | 0.47 (0.30–0.64) | 0.38 (0.21–0.55) | 0.09 (−0.15–0.34) | 0.457 | 0.08 (−0.16–0.32) | 0.519 |
BMI (kg/m2) | −0.01 (−0.04–0.02) | −0.02 (−0.05–0.01) | 0.01 (−0.04–0.05) | 0.714 | 0.03 (−0.02–0.07) | 0.282 |
Change within 12 months | ||||||
Alcohol consumption (g/week) | −3.43 (−7.81–0.95) | −2.75 (−7.15–1.65) | −0.68 (−6.89–5.53) | 0.830 | 0.28 (−6.25–6.81) | 0.933 |
Physical activity (MET-hour/day) | 0.43 (0.25–0.62) | 0.14 (−0.04–0.32) | 0.29 (0.03–0.55) | 0.027 | 0.27 (0.01–0.52) | 0.038 |
BMI (kg/m2) | 0.03 (−0.01–0.06) | 0.02 (−0.02–0.06) | 0.01 (−0.04–0.06) | 0.679 | 0.01 (−0.04–0.07) | 0.634 |
. | Intervention group (n = 2,947) . | Control group (n = 2,944) . | Difference in change (Intervention group − Control group) . | P . | Adjusted difference of change (Intervention group − Control group) . | P . |
---|---|---|---|---|---|---|
Change within 6 months | ||||||
Alcohol consumption (g/week) | −2.75 (−6.33–0.82) | −5.25 (−9.45 to −1.06) | 2.50 (−3.01–8.01) | 0.375 | 2.59 (−3.22–8.40) | 0.382 |
Physical activity (MET-hour/day) | 0.47 (0.30–0.64) | 0.38 (0.21–0.55) | 0.09 (−0.15–0.34) | 0.457 | 0.08 (−0.16–0.32) | 0.519 |
BMI (kg/m2) | −0.01 (−0.04–0.02) | −0.02 (−0.05–0.01) | 0.01 (−0.04–0.05) | 0.714 | 0.03 (−0.02–0.07) | 0.282 |
Change within 12 months | ||||||
Alcohol consumption (g/week) | −3.43 (−7.81–0.95) | −2.75 (−7.15–1.65) | −0.68 (−6.89–5.53) | 0.830 | 0.28 (−6.25–6.81) | 0.933 |
Physical activity (MET-hour/day) | 0.43 (0.25–0.62) | 0.14 (−0.04–0.32) | 0.29 (0.03–0.55) | 0.027 | 0.27 (0.01–0.52) | 0.038 |
BMI (kg/m2) | 0.03 (−0.01–0.06) | 0.02 (−0.02–0.06) | 0.01 (−0.04–0.06) | 0.679 | 0.01 (−0.04–0.07) | 0.634 |
Note: Data are mean (95% CI). Adjusted difference of change is adjusted for age, sex, number of healthy lifestyle habits, clinical history, health literacy, marital status, education, occupation, employment status, household income, and usage history of the cancer risk prediction tool in a smartphone app.
Abbreviations: MET, metabolic equivalent; BMI, body mass index; CI, confidence interval.
In the subgroup analyses, there were no statistically significant differences in the improvement in lifestyle between the groups, except for participants without a spouse, who were more likely to improve their lifestyle within 12 months in the intervention group than in the control group (20.7% vs. 16.2%; OR, 1.34; 95% CI, 1.05–1.73; P = 0.020; Supplementary Tables S1 and S2). Of note, the proportion of participants who subscribed to the lifestyle modification program was significantly higher in the intervention group than in the control group among participants with low health literacy (28.8% vs. 22.3%; OR, 1.41: 95% CI, 1.15–1.72; P < 0.001; Table 4), or with two or fewer healthy lifestyle habits (22.4% vs. 14.8%; OR, 1.67; 95% CI, 1.24–2.25; P < 0.001). In the additional post hoc analyses, 709 of 784 (90.4%) and 473 of 730 (64.8%) subscribers of the lifestyle modification program chose the course related to the most influential lifestyle habit for preventing cancer in the intervention and control groups, respectively.
. | Intervention group (n = 2,947) . | Control group (n = 2,944) . | OR (95% CI) . | P . | Adjusted OR (95% CI) . | P . |
---|---|---|---|---|---|---|
Age | ||||||
40–49 y | 364/1,326 (27.5%) | 326/1,326 (24.6%) | 1.16 (0.98–1.38) | 0.093 | 1.14 (0.95–1.38) | 0.161 |
50–59 y | 320/1,259 (25.4%) | 309/1,257 (24.6%) | 1.05 (0.87–1.25) | 0.629 | 1.02 (0.84–1.24) | 0.829 |
60–64 y | 100/362 (27.6%) | 92/361 (25.5%) | 1.12 (0.80–1.55) | 0.515 | 0.94 (0.63–1.42) | 0.786 |
Sex | ||||||
Male | 353/1,565 (22.6%) | 312/1,563 (20.0%) | 1.17 (0.98–1.39) | 0.076 | 1.11 (0.92–1.33) | 0.271 |
Female | 431/1,382 (31.2%) | 415/1,381 (30.1%) | 1.05 (0.90–1.24) | 0.517 | 1.03 (0.86–1.24) | 0.725 |
Health literacy | ||||||
Low | 297/1,030 (28.8%) | 233/1,043 (22.3%) | 1.41 (1.15–1.72) | <0.001 | 1.36 (1.09–1.69) | 0.007 |
Medium | 252/1,028 (24.5%) | 266/1,004 (26.5%) | 0.90 (0.74–1.10) | 0.306 | 0.91 (0.73–1.14) | 0.423 |
High | 235/885 (26.6%) | 228/891 (25.6%) | 1.05 (0.85–1.30) | 0.643 | 0.96 (0.76–1.22) | 0.762 |
Clinical history | ||||||
None | 613/2,280 (26.9%) | 569/2,279 (25.0%) | 1.11 (0.97–1.26) | 0.139 | 1.08 (0.93–1.25) | 0.315 |
Any | 171/667 (25.6%) | 158/665 (23.8%) | 1.11 (0.86–1.42) | 0.427 | 1.03 (0.78–1.36) | 0.825 |
Marital status | ||||||
With a spouse | 551/2,119 (26.0%) | 510/2,103 (24.3%) | 1.10 (0.96–1.26) | 0.190 | 1.04 (0.89–1.21) | 0.624 |
Without a spouse | 230/818 (28.1%) | 215/826 (26.0%) | 1.11 (0.89–1.38) | 0.341 | 1.11 (0.88–1.41) | 0.383 |
Education | ||||||
High school or less | 261/1,300 (20.1%) | 262/1,312 (20.0%) | 1.01 (0.83–1.22) | 0.945 | 0.96 (0.77–1.19) | 0.699 |
College | 245/825 (29.7%) | 225/805 (28.0%) | 1.09 (0.88–1.35) | 0.436 | 1.10 (0.87–1.39) | 0.441 |
University or higher | 275/816 (33.7%) | 238/815 (29.2%) | 1.23 (1.00–1.52) | 0.050 | 1.18 (0.95–1.48) | 0.136 |
Occupation | ||||||
Unemployed or homemaker | 83/255 (32.5%) | 84/285 (29.5%) | 1.15 (0.80–1.66) | 0.440 | 0.69 (0.16–2.92) | 0.613 |
White collar | 581/2,127 (27.3%) | 542/2,130 (25.4%) | 1.10 (0.96–1.26) | 0.166 | 1.09 (0.95–1.25) | 0.243 |
Blue collar | 85/447 (19.0%) | 86/432 (19.9%) | 0.94 (0.68–1.32) | 0.738 | 0.94 (0.66–1.33) | 0.724 |
Employment status | ||||||
Permanent | 445/1,744 (25.5%) | 420/1,736 (24.2%) | 1.07 (0.92–1.25) | 0.367 | 1.06 (0.91–1.25) | 0.441 |
Fixed-term | 202/707 (28.6%) | 167/651 (25.7%) | 1.16 (0.91–1.47) | 0.227 | 1.10 (0.85–1.41) | 0.468 |
Self-employment or administrator | 61/280 (21.8%) | 62/306 (20.3%) | 1.10 (0.74–1.63) | 0.651 | 0.97 (0.63–1.50) | 0.898 |
Income | ||||||
<3 million yen | 133/532 (25.0%) | 127/511 (24.9%) | 1.01 (0.76–1.33) | 0.956 | 0.93 (0.67–1.28) | 0.648 |
3–6 million yen | 297/1,157 (25.7%) | 250/1,086 (23.0%) | 1.15 (0.95–1.40) | 0.144 | 1.12 (0.91–1.39) | 0.291 |
6–9 million yen | 199/751 (26.5%) | 206/792 (26.0%) | 1.03 (0.82–1.29) | 0.828 | 1.00 (0.78–1.28) | 0.982 |
≥9 million yen | 150/486 (30.9%) | 142/532 (26.7%) | 1.23 (0.93–1.61) | 0.142 | 1.19 (0.88–1.59) | 0.256 |
Usage history of a smartphone app | ||||||
Yes | 81/243 (33.3%) | 65/199 (32.7%) | 1.03 (0.69–1.54) | 0.882 | 1.02 (0.65–1.60) | 0.932 |
No | 703/2,704 (26.0%) | 662/2,745 (24.1%) | 1.11 (0.98–1.25) | 0.109 | 1.07 (0.93–1.22) | 0.339 |
Number of healthy lifestyle habits | ||||||
≤2 | 134/597 (22.4%) | 90/610 (14.8%) | 1.67 (1.24–2.25) | <0.001 | 1.53 (1.11–2.10) | 0.009 |
3 | 309/1,191 (25.9%) | 308/1,180 (26.1%) | 0.99 (0.83–1.19) | 0.931 | 0.92 (0.75–1.13) | 0.424 |
4 | 341/1,159 (29.4%) | 329/1,154 (28.5%) | 1.05 (0.87–1.25) | 0.629 | 1.05 (0.86–1.28) | 0.635 |
. | Intervention group (n = 2,947) . | Control group (n = 2,944) . | OR (95% CI) . | P . | Adjusted OR (95% CI) . | P . |
---|---|---|---|---|---|---|
Age | ||||||
40–49 y | 364/1,326 (27.5%) | 326/1,326 (24.6%) | 1.16 (0.98–1.38) | 0.093 | 1.14 (0.95–1.38) | 0.161 |
50–59 y | 320/1,259 (25.4%) | 309/1,257 (24.6%) | 1.05 (0.87–1.25) | 0.629 | 1.02 (0.84–1.24) | 0.829 |
60–64 y | 100/362 (27.6%) | 92/361 (25.5%) | 1.12 (0.80–1.55) | 0.515 | 0.94 (0.63–1.42) | 0.786 |
Sex | ||||||
Male | 353/1,565 (22.6%) | 312/1,563 (20.0%) | 1.17 (0.98–1.39) | 0.076 | 1.11 (0.92–1.33) | 0.271 |
Female | 431/1,382 (31.2%) | 415/1,381 (30.1%) | 1.05 (0.90–1.24) | 0.517 | 1.03 (0.86–1.24) | 0.725 |
Health literacy | ||||||
Low | 297/1,030 (28.8%) | 233/1,043 (22.3%) | 1.41 (1.15–1.72) | <0.001 | 1.36 (1.09–1.69) | 0.007 |
Medium | 252/1,028 (24.5%) | 266/1,004 (26.5%) | 0.90 (0.74–1.10) | 0.306 | 0.91 (0.73–1.14) | 0.423 |
High | 235/885 (26.6%) | 228/891 (25.6%) | 1.05 (0.85–1.30) | 0.643 | 0.96 (0.76–1.22) | 0.762 |
Clinical history | ||||||
None | 613/2,280 (26.9%) | 569/2,279 (25.0%) | 1.11 (0.97–1.26) | 0.139 | 1.08 (0.93–1.25) | 0.315 |
Any | 171/667 (25.6%) | 158/665 (23.8%) | 1.11 (0.86–1.42) | 0.427 | 1.03 (0.78–1.36) | 0.825 |
Marital status | ||||||
With a spouse | 551/2,119 (26.0%) | 510/2,103 (24.3%) | 1.10 (0.96–1.26) | 0.190 | 1.04 (0.89–1.21) | 0.624 |
Without a spouse | 230/818 (28.1%) | 215/826 (26.0%) | 1.11 (0.89–1.38) | 0.341 | 1.11 (0.88–1.41) | 0.383 |
Education | ||||||
High school or less | 261/1,300 (20.1%) | 262/1,312 (20.0%) | 1.01 (0.83–1.22) | 0.945 | 0.96 (0.77–1.19) | 0.699 |
College | 245/825 (29.7%) | 225/805 (28.0%) | 1.09 (0.88–1.35) | 0.436 | 1.10 (0.87–1.39) | 0.441 |
University or higher | 275/816 (33.7%) | 238/815 (29.2%) | 1.23 (1.00–1.52) | 0.050 | 1.18 (0.95–1.48) | 0.136 |
Occupation | ||||||
Unemployed or homemaker | 83/255 (32.5%) | 84/285 (29.5%) | 1.15 (0.80–1.66) | 0.440 | 0.69 (0.16–2.92) | 0.613 |
White collar | 581/2,127 (27.3%) | 542/2,130 (25.4%) | 1.10 (0.96–1.26) | 0.166 | 1.09 (0.95–1.25) | 0.243 |
Blue collar | 85/447 (19.0%) | 86/432 (19.9%) | 0.94 (0.68–1.32) | 0.738 | 0.94 (0.66–1.33) | 0.724 |
Employment status | ||||||
Permanent | 445/1,744 (25.5%) | 420/1,736 (24.2%) | 1.07 (0.92–1.25) | 0.367 | 1.06 (0.91–1.25) | 0.441 |
Fixed-term | 202/707 (28.6%) | 167/651 (25.7%) | 1.16 (0.91–1.47) | 0.227 | 1.10 (0.85–1.41) | 0.468 |
Self-employment or administrator | 61/280 (21.8%) | 62/306 (20.3%) | 1.10 (0.74–1.63) | 0.651 | 0.97 (0.63–1.50) | 0.898 |
Income | ||||||
<3 million yen | 133/532 (25.0%) | 127/511 (24.9%) | 1.01 (0.76–1.33) | 0.956 | 0.93 (0.67–1.28) | 0.648 |
3–6 million yen | 297/1,157 (25.7%) | 250/1,086 (23.0%) | 1.15 (0.95–1.40) | 0.144 | 1.12 (0.91–1.39) | 0.291 |
6–9 million yen | 199/751 (26.5%) | 206/792 (26.0%) | 1.03 (0.82–1.29) | 0.828 | 1.00 (0.78–1.28) | 0.982 |
≥9 million yen | 150/486 (30.9%) | 142/532 (26.7%) | 1.23 (0.93–1.61) | 0.142 | 1.19 (0.88–1.59) | 0.256 |
Usage history of a smartphone app | ||||||
Yes | 81/243 (33.3%) | 65/199 (32.7%) | 1.03 (0.69–1.54) | 0.882 | 1.02 (0.65–1.60) | 0.932 |
No | 703/2,704 (26.0%) | 662/2,745 (24.1%) | 1.11 (0.98–1.25) | 0.109 | 1.07 (0.93–1.22) | 0.339 |
Number of healthy lifestyle habits | ||||||
≤2 | 134/597 (22.4%) | 90/610 (14.8%) | 1.67 (1.24–2.25) | <0.001 | 1.53 (1.11–2.10) | 0.009 |
3 | 309/1,191 (25.9%) | 308/1,180 (26.1%) | 0.99 (0.83–1.19) | 0.931 | 0.92 (0.75–1.13) | 0.424 |
4 | 341/1,159 (29.4%) | 329/1,154 (28.5%) | 1.05 (0.87–1.25) | 0.629 | 1.05 (0.86–1.28) | 0.635 |
Note: Data are number (%), unless otherwise specified. Adjusted OR is adjusted for age, sex, number of healthy lifestyle habits, clinical history, health literacy, marital status, education, occupation, employment status, household income, and usage history of the cancer risk prediction tool in a smartphone app.
Abbreviations: OR, odds ratio; CI, confidence interval.
Discussion
In this study, we showed that paper-based personalized advice based on a prediction model of 10-year cumulative cancer risk had no statistically significant effect in improving lifestyle compared to standard advice for cancer prevention not based on a prediction model among a general population. However, the personalized advice did motivate participants with low health literacy or with two or fewer healthy lifestyle habits to subscribe to the lifestyle modification program.
Participants in the intervention group were given information based on a cancer risk prediction model, including (i) the probability of cancer occurrence; (ii) a recommended course of a lifestyle modification program to improve the most influential unhealthy lifestyle habit for preventing cancer; and (iii) the probability of cancer occurrence when the most influential lifestyle habit was improved, as personalized advice as opposed to standard advice. However, this information did not motivate participants to modify their lifestyle (Table 2). On the basis of a previous study (26), we speculated that personalized advice which included messages based on behavior change techniques, such as changing risk perception or feelings associated with adopting behaviors, would induce participants to subscribe to a lifestyle modification program and then change their unhealthy lifestyle. We also speculated that a personalized achievable goal in modifying lifestyle might be more effective than a common goal based on the threshold of the prediction model, because the difficulty of achieving healthy lifestyle habits varies by the current lifestyle and the type of target lifestyle aimed for. On the other hand, among our participants with low health literacy who had difficulty in interpreting health messages (27), the proportion of subscribers to the lifestyle modification program significantly increased when provided with data on their cancer risk (Table 4). Provision of a single appropriate option as the recommended course to participants with low health literacy might have been a cue to modification of their unhealthy lifestyle habits, even if they found it difficult to choose from among multiple options (28). In addition, the number of subscribers to the “quitting smoking” course among participants with personalized advice was more than twice that of participants with the standard advice (Table 2). Furthermore, while more than 90% of subscribers to the lifestyle modification program in the intervention group chose the recommended course based on the most influential lifestyle habit for preventing cancer in the personalized advice, only 64.8% of subscribers in the control group chose the course which they would have been recommended if they had been in the intervention group. Therefore, information based on the cancer risk prediction model might facilitate selection of an appropriate option.
Provision of information on general cancer risk factors along with personalized cancer risk based on the prediction model might aid the dissemination of knowledge on cancer prevention. Previous studies have shown poor awareness of the role of physical activity in preventing cancer (29, 30) and low adherence to adequate physical activity in the Japanese population (8). Although approximately 70% of participants in this study did not obtain adequate physical activity, the highest among the lifestyle habits investigated, the number of participants who subscribed to the “being more active” course was largest (Tables 1 and 2). In addition, participants with two or fewer healthy lifestyle habits in the intervention group more frequently subscribed to the lifestyle modification program than those in the control group (Table 4). Calculating cancer risk using a prediction model might be an opportunity to communicate evidence-based information on general cancer risk factors. Despite the relatively high awareness of smoking as a cancer risk factor (29); however, the rate of reduction in smoking is decreasing in Japan (31), which suggested that it is difficult to modify lifestyle simply by communicating information on risk factors. One reason for this is that health behavior is influenced at multiple levels—individual, social, environmental, and policy—a characteristic proposed in the ecological model (32). Further studies on appropriate interventions to communicate knowledge of cancer risk factors and modification of lifestyle using the cancer risk prediction model are needed.
This is the first randomized controlled trial to investigate the effect of paper-based personalized advice based on cancer risk estimated by a prediction model on improvement of multiple lifestyle habits among a geographically widespread general population in Japan. However, this study has some limitations. First, only 10.9% (6,534 of 59,892) of individuals who were sent the consent form and baseline questionnaire were enrolled (Fig. 1). Therefore, participants might be more health-conscious than the general population. In fact, the proportion of participants with two or fewer healthy lifestyle habits in this study was lower than that in the study used to develop the prediction model of 10-year cumulative cancer risk (8, 11). Such possible self-selection bias may limit the generalizability of the results (33, 34). Second, 21.1% (621 of 2947) of participants in the intervention group and 20.1% (592 of 2,944) in the control group did not answer the 6-month questionnaire used to evaluate the primary outcome (Fig. 1). Although the lifestyle changes of participants who did not answer the 6-month questionnaire were not known, we analyzed the primary outcome on the conservative assumption that participants who did not respond to this questionnaire did not change their lifestyle. In addition, we also analyzed participants who did answer the 6-month questionnaire as a sensitivity analysis. In both analyses, there were no significant differences between the intervention and control groups. Third, participants in the control group could learn their probability of cancer occurrence by using the cancer risk prediction tool provided through our website or the company's smartphone app independently of the study, albeit that we did not inform them of the availability of these tools, and this might have biased the estimated effect of the personalized advice toward the null as contamination bias. However, the personalized advice had no effect on the primary or secondary outcomes among participants without a usage history for the app (Table 4; Supplementary Tables S1 and S2). Fourth, information on lifestyle was collected by self-reported questionnaire. Although the questionnaires were based of self-administered questionnaires used in a population-based cohort study in Japan (19), whose data were used to develop the prediction model of 10-year cumulative cancer risk (8), misclassification of lifestyle habits was unavoidable due to over- or under-reporting to meet socially desirable behavior and recall bias (35–39). Thus, misclassification due to social desirability bias might have biased the estimated effect of the personalized advice toward the null. Digital tools offer promise for the collection of accurate information on lifestyle habits, especially physical activity (40). Fifth, the effectiveness of personalized advice on improvement in lifestyle in cancer survivors was unknown because the number of cancer survivors enrolled in this study was small. Only 56 (1.9%) participants in the intervention group and 48 (1.6%) in the control group had a clinical history of cancer (Table 1). Further study is needed to evaluate the effectiveness of personalized advice for cancer survivors.
In this study, we found no statistically significant effect of personalized advice based on a prediction model for 10-year cumulative cancer risk on improvement in lifestyle compared with standard advice for cancer prevention not based on a prediction model among a general population. However, the personalized advice might have motivated individuals with low health literacy and those adhering to two or fewer of five healthy lifestyle habits to modify their unhealthy lifestyle. Further research on the use of cancer risk prediction models in improving lifestyle is needed.
Authors' Disclosures
K. Yuwaki reports personal fees from The Dai-ichi Life Insurance Company, Limited during the conduct of the study. A. Kuchiba reports grants from The Japan Agency for Medical Research and Development during the conduct of the study, as well as personal fees from Chugai Pharmaceutical Co., Ltd outside the submitted work. A. Otsuki reports grants from The Japan Agency for Medical Research and Development during the conduct of the study. M. Odawara reports grants from The Japan Agency for Medical Research and Development during the conduct of the study. M. Inoue reports grants from The Japan Agency for Medical Research and Development during the conduct of the study. S. Tsugane reports grants from The Japan Agency for Medical Research and Development during the conduct of the study. T. Shimazu reports grants from The Japan Agency for Medical Research and Development during the conduct of the study, as well as grants from The Dai-ichi Life Insurance Company, Limited outside the submitted work. No disclosures were reported by the other authors.
Authors' Contributions
K. Yuwaki: Conceptualization, formal analysis, methodology, writing–original draft. A. Kuchiba: Formal analysis, methodology, writing–review and editing. A. Otsuki: Writing–review and editing. M. Odawara: Writing–review and editing. T. Okuhara: Writing–review and editing. H. Ishikawa: Writing–review and editing. M. Inoue: Writing–review and editing. S. Tsugane: Writing–review and editing. T. Shimazu: Conceptualization, supervision, funding acquisition, methodology, writing–review and editing.
Acknowledgments
We are grateful to Hanako Saito, Yuko Miyamoto, and Yoko Hirose for their cooperation with this study as research staff. We also thank Kenji Nishiiri, Mika Yamanaka, and other members at the Underwriting and Medical Department, The Dai-ichi Life Insurance Company, Limited, for supporting the recruitment of participants. The Japan Agency for Medical Research and Development had no role in the study design, data collection, analysis, interpretation, manuscript drafting, or decision to submit the manuscript for publication. This study was supported by the Japan Agency for Medical Research and Development under grant number JP19ck0106270.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.