Article Text

Original research
Effectiveness of a targeted primary preventive intervention in a high-risk group identified using an efficiency score from data envelopment analysis: a randomised controlled trial of local residents in Japan
  1. Sho Nakamura1,2,3,
  2. Satoru Kanda2,3,4,
  3. Hiroko Endo5,
  4. Emiko Yamada6,
  5. Miki Kido6,
  6. Shoko Sato6,
  7. Iku Ogawa6,
  8. Rina Inoue3,
  9. Masanori Togashi6,
  10. Ken Izumiya7,
  11. Hiroto Narimatsu1,2,3,8
  1. 1Graduate School of Health Innovation, Kanagawa University of Human Services, Kawasaki, Kanagawa, Japan
  2. 2Cancer Prevention and Control Division, Kanagawa Cancer Center Research Institute, Yokohama, Kanagawa, Japan
  3. 3CIKOP, Specified Nonprofit Corporation, Yamagata, Yamagata, Japan
  4. 4Office of Health Policy, Department of Health and Welfare, Iwate Prefectural Government Office, Morioka, Iwate, Japan
  5. 5Section of Welfare and Child Service, Takahata Town Office, Takahata, Higashiokitama-gun, Yamagata, Japan
  6. 6Section of Health and Longevity Service, Takahata Town Office, Takahata, Higashiokitama-gun, Yamagata, Japan
  7. 7Division of Urology, Takahata Public Hospital, Takahata, Higashiokitama-gun, Yamagata, Japan
  8. 8Department of Genetic Medicine, Kanagawa Cancer Center, Yokohama, Kanagawa, Japan
  1. Correspondence to Dr Sho Nakamura; research{at}nakasho.org

Abstract

Objective To determine whether a minimal intervention based on the data envelopment analysis (DEA)-identified efficiency score effectively prevents hypertension.

Design Randomised controlled trial.

Setting Takahata town (Yamagata, Japan).

Participants Residents aged 40–74 years belonged to the information provision group for specific health guidance. Participants with a blood pressure ≥140/90 mm Hg, those taking antihypertensive medication, or those with a history of cardiac diseases were excluded. Participants were consecutively assigned based on their health check-up visit at a single centre from September 2019 to November 2020 and were followed up at the check-up in the following year, until 3 December 2021.

Intervention A targeted approach using minimal intervention. Target was identified using DEA and 50% of participants with higher risk were targeted. The intervention was notifying the results of their risk of hypertension according to the efficiency score obtained by the DEA.

Primary outcome measures A reduction in the proportion of participants who developed hypertension (≥140/90 mm Hg or taking antihypertensive medication).

Results A total of 495 eligible participants were randomised, and follow-up data were available for 218 and 227 participants in the intervention and control groups, respectively. The risk difference for the primary outcome was 0.2% (95% CI −7.3 to 6.9) with 38/218 (17.4%) and 40/227 (17.6%) events in the intervention and control group, respectively (Pearson’s χ2 test, p=0.880). The adjusted OR of the effect of the intervention was 0.95 (95% CI 0.56 to 1.61, p=0.843), and that of the efficiency score (10-rank increase) was 0.81 (95% CI 0.74 to 0.89, p<0.0001).

Conclusions Minimal intervention to a high-risk population stratified by DEA was not effective in reducing the onset of hypertension in 1 year. The efficiency score could predict the risk of hypertension.

Trial registration number UMIN000037883

  • hypertension
  • preventive medicine
  • risk management

Data availability statement

Data are available upon reasonable request. The datasets generated and analysed during the current study are not publicly available for ethical reasons, but individual deidentified participant data (including data dictionaries) will be available upon approval of the protocols, including the purpose of data sharing (e.g., verification of reproducibility to prevent research misconduct), by the Research Ethics Review Board at the Graduate School of Health Innovation, Kanagawa University of Human Services, and at Takahata Town. We cannot provide the datasets for the purpose of secondary use in research, as consent for such use has not been obtained. Study protocol is available as a Supplement.

http://creativecommons.org/licenses/by-nc/4.0/

This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.

Statistics from Altmetric.com

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

STRENGTHS AND LIMITATIONS OF THIS STUDY

  • Single-blinded randomised controlled trial.

  • The efficiency score was calculated using data envelopment analysis.

  • The calculated efficiency score was used as a risk score to identify high-risk populations among normotensive individuals for the intervention.

  • The intervention was mild, and we did not assess whether the intervention was for the participants.

Introduction

High blood pressure decreases the life expectancy or disability-adjusted life years, owing to the increased risk of cardiovascular and cerebrovascular diseases.1 High blood pressure and hypertension are preventable disease, with ample evidence of interventions for the prevention and management of hypertension. A guideline from the American College of Cardiology/American Heart Association in 2017 recommended non-pharmacologic interventions, such as weight reduction, salt-intake reduction, healthy eating, physical activity and moderate alcohol intake, for individuals with elevated blood pressure (120–129/<80 mm Hg) to prevent the development of hypertension.2 3

Primary prevention, aiming to prevent the onset of hypertension, could be achieved by the abovementioned non-pharmacological interventions.4 It is essential for reducing the burden caused by hypertension in individuals with normal (<120/<80 mm Hg) to high blood pressure stage 1 (120–140/<90 mm Hg). However, at present, these interventions have limited effects. One reason for the lack of effectiveness could be that the current primary prevention adopts a classical one-size-fits-all approach, wherein the context is limited to a general extent that is insufficient to motivate or induce behavioural change. Furthermore, it is difficult to adhere to all of the recommendations, while the optimal intervention that should be prioritised for each individual is not recognised. Personalised interventions that are prioritised according to the individuals’ risk could be expected to provide more effective primary prevention. One approach to inculcate individualised preventive interventions is to adapt recent findings from genomic research on hypertension, such as genome-wide association studies or studies that predict future morbidity based on genetic risk.5 6

Nonetheless, genomic information has not yet been widely applied in primary prevention. Furthermore, the predictive ability of genomic data to determine the future morbidity of hypertension is unsatisfactory because of many misclassifications.7 Moreover, although the genotyping cost has decreased, the cost remains extravagant in a population-level approach. We anticipate that data envelopment analysis (DEA) could enable personalised primary prevention because of two characteristic advantages. First, unlike in a parametric model, such as linear regressions assuming that the single regression equation (mean) obtained through a single optimisation fits the whole, the DEA is a non-parametric model that measures the efficiency of each observation through the optimisation of every single observation. Second, an efficiency score from the DEA could serve as a tool to distinguish people with a high risk of future disease from the normotensive population.8

DEA, which is renowned in the field of management engineering, uses input(s) and output(s) in a linear programming model to evaluate the efficiency of the decision-making unit (DMU)—that is, for each observation. The DMU with a maximum output (eg, profit) and a minimum input (eg, investment) is the most efficient unit in the observed set.9 10 All DMUs other than the efficient DMUs are inefficient, and the amount of inefficiency can be calculated. The application of DEA in the evaluation of disease risk is based on a concept that uses the above-described inefficiency as a measure of disease risk. We have previously reported that calculating an efficiency score using DEA in a normotensive population allowed us to identify individuals at high risk of developing hypertension. This result was replicated in another population.7 11 However, both of the abovementioned studies were based on a cohort study, and, therefore, the effect of targeting high-risk individuals, according to the efficiency score, for providing primary prevention is unknown. Thus, we designed an interventional study to explore the potential usefulness of the DEA in community healthcare, especially regarding primary prevention.

This randomised controlled trial (RCT) aims to test our hypothesis that targeting individuals with a high risk of developing hypertension, based on the DEA-identified efficiency score, through a minimal intervention of risk notification is effective for the prevention of hypertension.

Methods

Study design

This single-centre, stratified (by sex and use of lifestyle measures to prevent hypertension), blocked (block size of four and six within a strata), single-blind, parallel-group, 1:1 RCT in Japan was conducted to compare the prevalence of incident hypertension between participants who were notified of their DEA-calculated risk of hypertension and those who were not. Written informed consent was obtained from all participants. This research was performed in accordance with the Declaration of Helsinki. This trial was registered in the UMIN Clinical Trials Registry (UMIN-CTR); 1 September 2019. A protocol amendment concerning the recruitment period was made due to the insufficient number of participants owing to the short initial recruitment period (see online supplemental file 1).

Study population

We recruited local residents, aged 40–74 years, from Takahata town who underwent the specific health check-up (Tokuteikenshin: the mandatory annual health check-up in Japan)12 at a single centre (Genkikan: a facility of the Takahata town office with a department of health that was situated next to Takahata Public Hospital). Takahata is in the Yamagata prefecture, located approximately 300 km north of Tokyo, Japan. Potential participants for the information provision (one of the specific health guidance groups classified by the result of specific health check-ups) were screened for eligibility.12 We excluded participants who had a blood pressure (systolic/diastolic) ≥140/90 mm Hg, were taking antihypertensive medication or had a history of cardiac disease (angina pectoris, myocardial infarction, etc). Recruitment for the study was conducted from September 2019 to November 2020, and all participants were followed up at the next year’s health check-up until 3 December 2021. The time point of the follow-up was once. The flow diagram of the participant selection process for this study is shown in figure 1. Among the 1286 potential participants who consented to participate in the study, 495 eligible participants underwent randomisation, and data from 247 and 248 participants, in the intervention and control groups, respectively, were included in the intention-to-treat (ITT) analysis. In the full analysis set (FAS), we excluded participants lost-to-follow-up, including three participants in each group who were ineligible for specific health check-ups due to the age cut-off. Thus, data from 218 and 227 participants, in the intervention and control groups, respectively, were included in the FAS.

Figure 1

Flow diagram of the participant selection process. *Angina pectoris and myocardial infarction.

Outcome measures

The primary outcome was the reduction in the proportion of participants who developed hypertension, defined as any of the following conditions at the health check-up in the following year: systolic blood pressure ≥140 mm Hg, diastolic blood pressure ≥90 mm Hg or taking antihypertensive medication. The secondary outcomes were the examination at the health check-up; onset rate of cardiovascular diseases, mortality rate, death related to cardiovascular diseases during 5 years from baseline; total medical expense of each individual in 5 years (sum of annual medical expenditure for 5 years after baseline recruitment) and difference of the annual medical expense between the baseline year and each 5 years after baseline recruitment.

Study intervention

Daily salt intake was calculated using Tanaka’s formula based on the serum sodium and creatinine as well as urinary sodium and creatinine from the spot urine sample and conveyed to all participants in both the intervention and control groups via a letter.13 DEA was performed as described previously,7 by including each participant in the producible set, which included data from the health check-up in 2018 and fulfilling the same eligibility criteria as in this analysis. However, physical activity, which was used as an input variable in the previous study, was not used as input in this study,7 although it was available. Physical activity is not routinely evaluated in healthcare, including the health check-up, although it has been recommended as a vital sign recently.14 15 We excluded physical activity to avoid a situation where even when the results of this study were positive, this technique could not be disseminated because of the unavailable physical activity data. We created the input-oriented constant returns-to-scale (CRS) Charnes-Cooper-Rhodes (CCR) model of DEA, wherein the input was the inverse of salt intake, and the outputs were the inverse of systolic and diastolic blood pressure. We used the CCR model to assess both technical and scale efficiency; the former reflects individuals’ constitution, including genetic background, while the latter reflects lifestyle, including salt intake. However, although the linear relationship between salt intake and blood pressure is well known, a strong assumption of CRS in the CCR model may create biased or inconsistent results; thus, we calculated the efficiency score using the variable returns-to-scale Banker-Charnes-Cooper (BCC) model of DEA as unprespecified sensitivity analysis.16 17 The efficiency scores were calculated using the dea function in the R Benchmarking (V.0.29) package.18 Higher scores indicate higher efficiency and, in this study, efficient participants had lower blood pressure despite a relatively higher salt intake. The efficiency score was converted into a ranking from 1 to 101, with a higher rank (first), indicating a greater risk of future onset of hypertension. The intervention was performed for the participants whose rank for the CCR model’s efficiency score was higher than the 50th rank in the intervention group. Only 50% of the participants were selected for intervention for two reasons: first, we were concerned that informing individuals that their risk of hypertension would be less than 50% of the population, which may create an illusion that they do not have to care about their health and induce a negative impact on health. Second, we wished to clarify the efficacy of a targeted approach using DEA since a population-level approach to intervene in the entire population with normal blood pressure may not be feasible. Mailing of the educational content was not evident to decrease participants’ blood pressure in a previous study.19 In this study, the intervention was to notify the risk based on the efficiency score along with their salt intake and educational content on hypertension using postal mail. Educational content was an explanation that excess salt intake worsens blood pressure and its aetiology; however, the method of salt intake reduction was not mentioned in the letter. This intervention was performed only once. Another intervention, such as lifestyle modification, was not conducted.

Sample size

Among the residents of Takahata Town who underwent specific health check-ups in 2017 and 2018, 18% of residents who met the inclusion criteria of this study in 2017 developed hypertension in the next year (obtained from the internal resource of Takahata Town Office). There is no study that assessed the effect of targeted intervention based on the efficiency score from DEA. Thus, we assumed that the intervention effect is relevant to 0.2 point of the DEA score, which would induce a 50% decrease (incidence of 8.7%) in the outcome based on the available evidence that the OR (90% CI (CI)) of 0.1-point increase in the efficiency score is 0.66 (0.55 to 0.78). Based on these assumptions, we would need a sample size of 281 patients per group, with a two-sided 5% significance level and a power of 80%, and given an anticipated dropout rate of 20%.

Randomisation

Eligible participants were randomly assigned to the intervention and control groups. Stratified permuted block randomisation was undertaken using a computer-generated random number list prepared by an investigator without involvement in the intervention. Participants were consecutively assigned in the order of their health check-ups and were blinded to the allocation.

Patient and public involvement statement

Takahata town is responsible for disseminating appropriate healthcare services to the residents. The need to identify high-risk residents among the normotensive population and the need for effective mild intervention were derived through communication and experience of health guidance provided to the public.20 Participants and/or the public were not involved in planning the design, performance, reporting or dissemination of this research.

Statistical analysis

All statistical analyses were performed with R (V.4.1.0) (R Core Team, Vienna, Austria).21 Differences in the characteristics at baseline and follow-up between the study arms were assessed using the Student’s t-test for continuous variables and the Pearson’s Χ2 test for categorical variables. We calculated Pearson’s product-moment correlation between the input variable used in the DEA and the rank of the efficiency score. We calculated the risk difference of the outcome and used Pearson’s Χ2 test to assess the difference in the proportion of the outcome. Furthermore, we used logistic regression analysis to further evaluate the effect of the intervention that was adjusted by the rank of the efficiency score, age, sex, BMI, smoking status, overall physical activity and the length of follow-up in the FAS. We checked the crude OR for each variable. The linearity of the continuous variable was graphically checked using the gam function in the R mgcv (V.1.8–38) package,22 and all continuous variables were linear in the logit. In addition, we checked multicollinearity using the vif function in the R car (V.3.0–12) package,23 and the highest value was 1.16, which indicated no concerning effect of multicollinearity. The area under the receiver operating characteristic curves (AUROC) was calculated using the roc function of the R pROC (V.1.18.0) package.24

We performed several unprespecified sensitivity analysis. Relative risk, risk difference and Pearson’s Χ2 test for the difference in the proportion of the outcome between the trial groups were assessed in four subgroups stratified by sex (males and females) and age (<mean, ≥mean). In addition, we performed a bootstrap DEA analysis of 2000 repeated procedures using dea.boot function in the R Benchmarking (V.0.29) package to obtain a bias-corrected efficiency score.18 25 We calculated Pearson’s product-moment correlation between the rank of these bias-corrected and raw DEA efficiency scores, the CCR model efficiency and the BCC model efficiency scores. We also assessed scale efficiency, which is calculated by dividing the efficiency score from the CCR model by that from the BCC model. Crude and adjusted ORs were calculated using the methods described above, with the CCR model efficiency score and scale efficiency interchanged.

Results

The baseline characteristics of the participants are listed in table 1, and there were no intergroup differences at baseline. The mean rank of the efficiency score was 55.7th (SD 30.2) in the intervention group and 55.9th (SD 29.6) in the control group. In the intervention group, 88/218 (40.4%) participants received the intervention (table 1 and figure 1). The mean daily salt intake for these 88 participants was 8.0 g/day (IQR 7.1–8.8), whereas for the other 130 participants was 10.6 g/day (IQR 9.7–11.7). The scatterplot of the ranking of the efficiency score and input factors in DEA are shown in figure 2 (salt intake) and in online supplemental eFigure 1 (systolic and diastolic blood pressures). The correlation coefficient between salt intake and the efficiency score was r(443)=0.85 (95% CI 0.82 to 0.87, p<0.0001). The correlation between the bias-corrected and raw DEA efficiency scores was r(443)=0.9996 (95% CI 0.9995 to 0.9996, p<0.0001) as shown in figure 3. Of the 176 (39.6%) participants whose rank of their raw DEA efficiency score was below 50, 3/176 (1.7%) participants’ rank became higher than 50 for the bias-corrected efficiency score. On the other hand, one participant’s rank fell below 50 for the bias-corrected efficiency score (1/269 (0.4%)).

Table 1

Baseline characteristics of the participants

Figure 2

Scatterplots of salt intake and efficiency score. The rank of the efficiency score is based on the efficiency score calculated using the data envelopment analysis, converted into a ranking from 1 to 101, with a higher rank (first) indicating an increased risk of hypertension. Salt intake was calculated using Tanaka’s method.12 The outcome event was incident hypertension (≥140/90 mm Hg or taking antihypertensive medication) during the follow-up period.

Figure 3

Scatter plots of bias-corrected and bias-uncorrected efficiency scores. The correlation coefficient was calculated using Pearson’s product–moment correlation.

The participants’ characteristics at follow-up are shown in table 2, which is also one of the secondary endpoints of the study. The lowest P-value for the difference of each characteristic between the intervention and control group was 0.024 for the length of follow-up, and none was below 0.001. We observed a difference of 21.2 days in the mean length of follow-up between the groups (P-value for difference=0.024).

Table 2

Characteristics of the participants at follow-up

The FAS was used for the main evaluation, since there were no data available after the randomisation for the participants who were lost to follow-up, thereby making us assess the results in the optimal condition. The risk difference was −0.2% (95% CI −7.3 to 6.9) with 38/218 (17.4%) and 40/227 (17.6%) outcome events for the intervention and control groups, respectively. The p value from the Pearson’s Χ2 test was 0.880. The relative risk among intervention group was 0.99. In the ITT analysis, which was performed as a sensitivity analysis, the outcome event was observed in 38/247 (15.4%) and 40/248 (16.1%) participants in the intervention and control groups, respectively. The risk difference was −0.7% (95% CI −7.2 to 5.7), and the p value from the Pearson’s Χ2 test was 0.917. Results of the sensitivity analysis are shown in eTable one in the Supplement.

The ORs for the outcome are shown in table 3. The adjusted OR of the effect of the intervention was 0.95 (95% CI 0.56 to 1.61, p=0.843). The AUROC for the multivariate logistic regression model was 0.72. For other variables that were adjusted, the adjusted ORs of the ranking of the efficiency score (10 rank increase) and age (1 year older) were 0.81 (95% CI 0.74 to 0.89, p<0.0001) and 1.09 (95% CI 1.04 to 1.31, p<0.0001), respectively. The AUROC for the efficiency score was 0.63, with a specificity of 0.63 and a sensitivity of 0.59.

Table 3

ORs of the onset of hypertension

The correlation between the technical and scale efficiency, which was calculated from the CCR model DEA, and pure technical efficiency, which was calculated from the BCC model DEA, were r(443)=0.982 (95% CI 0.979 to 0.985, p<0.0001) as shown in figure 4. The scale efficiency ranged from 0.69 to 1.00 with a median (IQR) of 0.98 (0.94 to 0.99). Of the 445, 19 (4.3%) participants’ rank of the efficiency score did not change between the models, 158 (35.5%) participants’ rank was higher (near 1, high risk) for the technical and scale efficiency, and 268 (58.9%) participants’ rank was higher for the pure technical efficiency. The ORs of the scale efficiency for the outcome are listed in online supplemental eTable 2. The adjusted OR of the scale efficiency (per 0.10 point increase) was 0.12 (95% CI 0.06 to 0.23, p<0.0001). The AUROC for the multivariate logistic regression model with scale efficiency was 0.78, with a specificity of 0.70 and a sensitivity of 0.75.

Figure 4

Scatter plots of efficiency scores from CCR and BCC model of DEA. The correlation coefficient was calculated using Pearson’s product–moment correlation. The rank of technical and scale efficiency indicates ranks based on efficiency score from the CCR model DEA, while the rank of pure technical efficiency indicates ranks based on efficiency score from the BCC model of DEA. Dashed grey line, a straight line with a slope of 1, is for reference. Scale efficiency was calculated by dividing the technical and scale efficiency by the pure technical efficiency. BCC, Banker-Charnes-Cooper; CCR, Charnes-Cooper-Rhodes; DEA, data envelopment analysis.

Discussion

In this RCT, the effect of a minimal intervention on the risk of hypertension was tested in a high-risk population. The results showed that sending a letter that notified normotensive participants (normal blood pressure on health check-ups) regarding their high risk of hypertension was insufficient to prevent the development of hypertension during a 1-year period.

A mild intervention was undertaken in this study: a risk ranking based on an efficiency score obtained from DEA, which was a rationale for the risk stratification, was sent via letter along with estimated salt intake to less than half of the participants in the intervention arm. The merits of the mild intervention were that it did not require effort or human resources and that adverse events were not anticipated. However, inducing a behavioural change in the participants was not enough to induce a positive result through the intervention.26 We assumed that participants whose results from the health check-up were within the normal range, which were at the ‘precontemplation’ stage of the transtheoretical model or the stages of change model. The intervention was anticipated to change their behaviour to the ‘contemplation’ stage and then voluntarily move to the ‘action’ stage through the ‘preparation’ stage.27 However, our results indicate that additional intervention is necessary to motivate them to step forward to the ‘preparation’ and then to the ‘action’ stage to prevent the development of hypertension. ‘Action’ refers to evidence-based non-pharmacologic interventions, such as salt reduction, healthy diet, weight reduction, smoking cessation and adequate physical activity.2 3 Our group has previously reported the long-term preventive effect of an intensive intervention that was performed by healthcare professionals.20 Given these points, future research that focuses on optimising the combination of the following aspects according to the available resource is needed; the intensity of an intervention, including its possible effect on the behavioural change and a strategy to identify the target population. In addition, the design of the intervention must be strengthened using an intervention based on appropriate theories of behaviour change.28 29

In addition, sending the ranking of the efficiency score along with the salt intake report could have hampered the motivation of the high-risk participants to make behavioural changes. As shown in figure 2, the ranking based on the efficiency score obtained from the DEA highly correlated with the estimated salt intake: the estimated salt intake was relatively lower for those who received the intervention letter with information regarding their ranking in the future risk of hypertension. As non-pharmacological interventions other than the salt reduction that were listed above were not mentioned in the letter, the intervening participants may have considered it difficult to reduce their salt intake and, thus, resigned themselves against acquiring a healthier lifestyle through salt reduction.26 Indeed, figure 2 shows the distribution of participants with lower salt intake who developed hypertension. One idea is to specify the target population for a more intensive intervention by selecting participants with a high risk according to the efficiency score and people with a lower intake of salt.

Another idea is to incorporate scale efficiency in target stratification. Our sensitivity analysis showed that scale efficiency allows for the prediction of future onset of hypertension in a normotensive population (eTable2). Higher scale inefficiency (1—scale efficiency) is anticipated for participants whose inefficiency was larger for the CCR model compared with the BCC model. In other words, inefficiency, including the effect of their lifestyle, calculated using the CCR model DEA, was larger compared with the pure technical inefficiency reflecting each individual’s constitution, including genetic and environmental factors, calculated using the BCC model DEA. Lifestyle changes are non-pharmacologic interventions that prevent the onset of hypertension. Therefore, risk stratification by scale efficiency is an alternative choice.

Furthermore, the short follow-up period might have limited us to observing predictive changes in the blood pressure, and a longer observation period may have enabled us to determine the benefits of a mild intervention. In this context, the sample size assumption might have been optimistic about capturing the risk reduction effect of hypertension during a 1-year period, and a larger sample size was required.

The strength of the study is that the efficiency score from DEA was validated for its predictive ability of the risk of hypertension among normotensive people in an RCT. A targeted approach for the high-risk population has benefits in terms of efficiency, which is improved with a tool that can accurately predict the latent risk.2 30 The efficiency score may serve as a tool to evaluate this latent risk, although the primary outcome of the RCT was not met. Thus, the novelty of our study that the effective high-risk primary preventive intervention strategy could be undertaken using the efficiency score was not evident. We have previously reported the predictive ability of the efficiency score from DEA by using data from a cohort study, and similar results are reported from a study in China.7 11 Compared with previous studies that were prone to errors in the data analysis due to the study design, which was not designed to detect the development of hypertension, the results of this study provide reliable data that adds to the evidence that an efficiency score can serve as a measure of disease risk. Additionally, the study indicates that the DEA technique could be applied to community healthcare, at least in primary prevention. Performing a window analysis comparing the efficiency score at baseline and at follow-up could have enabled us to assess the change in efficiency score and its difference between the trial groups. However, we could not perform this analysis since we were unable to calculate the efficiency score at follow-up due to data availability.

There are some limitations in our study. First, the study included individuals who came in for a health check-up. Thus, participants who developed hypertension during the follow-up period may have been more likely to not to come for a check-up in the following year because they would have started visiting the hospital for the treatment of hypertension. Second, the existence of an apprehension bias remains, although we stratified the participants according to whether they are taking any lifestyle measures to prevent hypertension from reducing allocation bias, since we anticipated that the intervention would cause behavioural change. Moreover, participants were blinded to the allocation to reduce the Hawthorne effect. Third, the analysis included in this paper is only limited to the primary endpoint and part of the secondary endpoint. Unable to incorporate the results regarding hypertensive disorders, information related to death, and medical expense is another limitation. Fourth, we must be cautious about the efficiency score, since the efficiency score calculated using the DEA is an optimal score within the given data set but not a perfect score for assessing a patient care outcome. Finally, we did not collect information on whether the participants actually read the letter that was sent and, thus, it is difficult to speculate about the extent to which the intervention reached the participants.

Conclusions

The primary endpoint of the effectiveness of minimal intervention aiming to prevent hypertension in a high-risk population stratified by DEA was not met. However, our study strengthens the evidence of the efficiency score from the DEA to serve as a tool to predict the future development of a hypertension.

Data availability statement

Data are available upon reasonable request. The datasets generated and analysed during the current study are not publicly available for ethical reasons, but individual deidentified participant data (including data dictionaries) will be available upon approval of the protocols, including the purpose of data sharing (e.g., verification of reproducibility to prevent research misconduct), by the Research Ethics Review Board at the Graduate School of Health Innovation, Kanagawa University of Human Services, and at Takahata Town. We cannot provide the datasets for the purpose of secondary use in research, as consent for such use has not been obtained. Study protocol is available as a Supplement.

Ethics statements

Patient consent for publication

Ethics approval

The study was approved by the Institutional Research Ethics Board at the Graduate School of Health Innovation, Kanagawa University of Human Services (approval number: HODAISHI-19-004), and Takahata-town (approval number: 2019-07-18). Written informed consent was obtained from all participants.

Acknowledgments

We thank all the staff of Takahata town and Takahata Public Hospital, and CIKOP, Specific Nonprofit Corporation, for their support of this study. The funder, Takahata town, Takahata Public Hospital and Iwate Prectural Government Office had no role in the design and conduct of the study, nor the decision to prepare and submit the manuscript for publication. We would like to thank Editage for the English language editing.

References

Supplementary materials

  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.

Footnotes

  • Contributors SN conceptualised and designed the study. SK and RI were responsible for randomisation and allocation of the study participants. HE, EY, MK, SS, IO, RI, MT and KI contributed to the data acquisition. SN analysed and interpreted the data and drafted the manuscript. HN critically revised the manuscript. All authors read and approved the final manuscript. HN are responsible for the overall contest as guarantor.

  • Funding This work was supported by JSPS KAKENHI Grant Number JP17K09152 to HN.

  • Competing interests CIKOP, Specific Nonprofit Corporation is contracted by Takahata town to provide health check-up reception services. The authors declare that they have no other competing interests.

  • Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.