Effect of Nudge-Based Intervention on Adherence to Physician Visit Recommendations and Early Health Outcomes among Individuals Identified with Chronic Kidney Disease in Screens

Study Design

We conducted an RCT to assess the effect of behavioral economics interventions on physician visits and early health outcomes among participants with CKD using the health system database. This RCT had three arms (the nudge-based letter, the clinical letter, and the control) to examine an add-on effect of the nudge-based letter compared with the clinical letter. The study protocol was published²¹ and registered in the University Hospital Medical Information Network Clinical Trial Registry (UMIN000035230).

Setting

The study participants were selected from beneficiaries of the nationwide employment-based health insurance plan in Japan (i.e., the national sample of employees of civil engineering and construction companies). In Japan, all adults over age 40 are required by law to undergo the national health screening program every year. The health system database includes annual health screening results and monthly measured medical claims data. The health screening results include demographics, body mass index, eGFR, urine protein, BP, hemoglobin A1c, medication use (antihypertensive, antidiabetic, and antihyperlipidemic drugs), and history of diseases (stroke and cardiovascular diseases). The claims data include diagnosis codes and procedure codes, which include information on all medical care provided and covered by the Japanese universal health insurance system.²²

Participant Selection

We selected study participants from a group of individuals who underwent health screening between April 2018 and March 2019 at ages 40–63 and were identified as having CKD (defined as an eGFR <60 ml/min per 1.74 m² or urine proteinuria [≥1+ by dipstick test]) during the screening. We excluded participants with a history of ESKD (eGFR<15 ml/min per 1.74 m², a history of receiving dialysis therapy, or a history of kidney transplants) (Figure 1). We estimated eGFR using the Japanese coefficient–modified Chronic Kidney Disease Epidemiology Collaboration equation, which has been validated using data on the Japanese population.²³ Urine dipstick is used to examine proteinuria in the Japanese health screening program. The institutional review board of Kyoto University approved all study procedures and waived informed consent for participants (approval no. C1420).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1.

Selection process for the study participants. The figure shows contents of the two types of letters, recommending physician visits.

Outcome Measures

The primary outcome was a visit to a physician within 6 months following the intervention. The visit was identified using medical claims data using diagnosis codes related to CKD (International Classification of Diseases, Tenth Revision codes [Supplemental Table 1]). The physician visit for this population is recommended by the guidelines developed by the Ministry of Health, Labor, and Welfare of Japan.²⁴ The secondary outcomes were eGFR, proteinuria, and systolic and diastolic BP 1 year after the intervention, measured in the following year’s health screening.

Intervention

We tested two types of interventions: nudge-based and clinical letters. Both letters were sent by mail to eligible participants with identified CKD. The letters recommended a visit to a physician to receive appropriate treatment for CKD and included a link to a website where participants could search for nearby health care providers. Both letters included information about the recipient’s results on eGFR, urine protein, and CKD-related risk factors (hypertension, diabetes, and smoking) identified during the screening. Figure 2 and Table 1 have details of the contents of the letters.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2.

Interventions with two types of letters. (A) Nudge-based letter. (B) Clinical letter. These letters were translated from Japanese into English. HbA1c, hemoglobin A1c.

Both the nudge-based and clinical letters provided the basic information (education component) about CKD (i.e., risk of cardiovascular disease and stroke due to CKD and risk of ESKD due to CKD). The difference between the two letters was that although the nudge-based letter used three nudge strategies developed in behavioral economics (loss aversion, default, and commitment), the clinical letter did not include them. Loss aversion emphasizes losses (progression to ESKD and incidence of cardiovascular diseases) if no action was taken by the recipient. The letters also described the inconveniences that participants would face in their daily lives if their CKD progressed to ESKD and required dialysis treatments. Given that people have the tendency to prefer avoiding losses to acquiring equivalent gains,²⁵ such loss-framed messaging has the potential to more effectively change participants’ behavior compared with focusing on the benefits of seeing a physician. We also provided the information on the clear action steps as the default option in order to avoid participants procrastinating because they are cognitively overwhelmed by too many options (e.g., which physician to see) to choose from (“choice overload”). Finally, a commitment device encourages participants to write down the name of the health care provider and the time of the physician visit in advance in the letter, which attempts to enforce their voluntarily imposed restrictions regarding their plan to visit a physician.

All three groups (two groups assigned to interventions and the control group) were informed of their results from the screening program (they received notification of their screening results, which did not include recommendations to visit physicians). The notification included information about test results, highlighting abnormal results.

Randomization

According to the timing of the screening, we delivered the interventions in two waves in order to ensure that intervention was received soon after undergoing the screening program: January 2019 (the first wave) and July 2019 (the second wave). After stratifyed by eGFR (≥60 or <60 ml/min), and urine protein (≥1+ or <1+), we randomized participants in a 2:2:1 ratio into the nudge-based letter group, the clinical letter group, and the control group, respectively, using a permuted block method with a block size of five for each stratum (Supplemental Figure 1).

Statistical Analyses

The sample sizes of 1700, 1700, and 850 patients for the nudge-based letter group, the clinical letter group, and the control group, respectively, were determined to achieve ≥80% power overall for all pairwise comparisons with the chi-squared test. The necessary sample sizes were calculated on the basis of a simulation under the assumption of the proportions of patients visiting physicians in each group at 0.25, 0.20, and 0.15, respectively.

The primary statistical analysis was performed according to the methods predetermined in the study protocol and trial registration shown in Supplemental Material. First, we described participant characteristics by groups. Also, to test whether participants’ characteristics were balanced after the randomization, we calculated pairwise standardized differences between assignment groups (clinical letter versus control and nudge-based letter versus control). Second, we examined the effect of the intervention on visits to physicians using an intention-to-treat analysis. For the analysis of the primary end point, we used logistic regression models adjusted for the allocation stratification variables (timing of the screening, eGFR, and urine protein). We calculated P values using an omnibus test (Wald test) to see whether there were any differences in outcomes (proportion of visits to physicians) between the three groups. A two-sided P value using the omnibus test of 0.05 was interpreted as a statistically significant difference in outcomes between the groups, and we proceeded to perform pairwise comparisons. We calculated adjusted proportions of physician visits for each group and their differences between groups (percentage point [pp]) using the model-based standardization technique, adjusting for the stratification variables (timing of the screening, eGFR, and urine protein).^26,27 Adjustment for the stratification variables in the primary analysis of the RCT is recommended to appropriately account for the study design.²⁸

As secondary outcomes, we also examined the effect of the interventions on early health outcomes (eGFR, proteinuria, and systolic and diastolic BP) 1 year after the intervention in the total population and in hypertensive participants (systolic BP ≥140 mm Hg, diastolic BP ≥90 mm Hg, or antihypertensive drug use). For continuous outcomes (eGFR and BP), we fitted ordinary least square regression models and estimated differences between groups, adjusting for the stratification variables (timing of the screening, eGFR, and urine protein). For binary outcomes (proteinuria), we used a logistic regression model and estimated adjusted proportion differences between groups (pp) using the marginal standardization technique, adjusting for the stratification variables. To account for multiple comparisons in health outcomes, we performed post hoc adjustments for P values by using the Benjamini–Hochberg procedure.²⁹

Secondary Analyses

We conducted several secondary analyses. First, to examine differences in the effect of interventions according to participants’ characteristics, we conducted stratified analyses by participants’ age, sex, coexistence of hypertension and diabetes, receipt of previous treatment, and baseline eGFR categories (<45, 45–59, and ≥60 ml/min). We conducted stratified analyses by diabetes, hypertension, age, sex, and previous treatment as prespecified in the study protocol. The previous treatment was defined as the presence of CKD treatment in the 6 months prior to the intervention by medical claims data. We also conducted the post hoc stratified analyses by baseline eGFR and health guidance intervention (we added these analyses during the analysis stage without seeing the actual data). We formally tested the interaction using the likelihood ratio test. Second, we tested for the possibility that observed changes in physician visits and early health outcomes may be explained by an increased likelihood of receiving a health guidance intervention after the health screening (participants who meet the eligibility criteria are required by law to receive health guidance provided by a trained instructor). Supplemental Material has additional details. Third, we reanalyzed the data additionally adjusting for the receipt of a health guidance intervention. Fourth, to account for missing health outcomes, we analyzed the data using weighted regression models using the inverse probability of observing the outcome variables as weights (observed individual-level variables were used to estimate the weights).³⁰ Among participants, 7.6% had missing values of health outcomes due to absence of second health screening results. Fifth, to address the possibility that some physician visits were due to health issues other than CKD, we reanalyzed the data using a narrower definition of physician visits: physician visits with the diagnosis of CKD (Supplemental Table 1) and claims related to the diagnostic tests for CKD (i.e., urine test for proteinuria and/or serum creatinine test). Sixth, to examine the effect of the interval between the initial screening and the intervention on the effects of intervention, we conducted stratified analysis by the intervals (<6, 6–9, and ≥10 months). Seventh, we conducted stratified analysis by waves (timing of the intervention). Eighth, we examined whether additionally adjusting for baseline characteristics (age, sex, body mass index, systolic BP, hemoglobin A1c, antihypertensive drugs, antidiabetic drugs, antihyperlipidemic drugs, and history of stroke and cardiovascular disease) affected our findings. Ninth, we reanalyzed the data, restricting to participants without previous CKD treatment. Finally, to investigate how much time it takes for a physician visit to affect health outcomes, we compared health outcomes of individuals on the basis of the time interval between a physician visit and the measurement of the follow-up health outcomes (<4 or ≥4 months).

All analyses were performed using Stata version 16.1 (StataCorp, College Station, TX) and SAS version 9.4 (SAS Institute).