Effectiveness of Specific Health Check-ups in Japan for the primary prevention of obesity-related diseases: a protocol for a target trial emulation

Abstract

Introduction

Concerns about public health threats have shifted towards obesity-related, non-communicable diseases in both developed and developing countries. Since 2008, Japan has adopted a nationwide universal screening programme called Specific Health check-ups (SHC) for the primary prevention of obesity-related, non-communicable diseases, namely, diabetes, hypertension and hyperlipidaemia. The effectiveness of SHC has not been systemically evaluated to date.

Methods and analysis

We will use the employment-based health insurance database, which includes both records of receipt versus non-receipt of SHC and healthcare claims. The study design adopts the target trial emulation framework to minimise the bias inherent to the observational study (eg, time-related bias and its consequences). The key components of trial emulation—eligibility criteria, treatment strategy, assignment procedure, follow-up period, outcome, causal contrast of interest and analysis plan—are detailed, along with the ideal target trial protocol. Briefly, we will conduct the nested-trial emulation approach that allows multiple trial entries. Persons aged 40–74 years will be eligible if they do not have a documented diagnosis of hypertension and diabetes and a history of receiving SHC at baseline. Participants will be classified according to the receipt or non-receipt of SHC service and followed for a maximum of 10 years until the withdrawal from the insurance plan, the outcome occurrence or the administrative censoring (the end of the available data, expected to be March 2022), whichever comes first. The 10-year incidence of diabetes/hypertension will be compared between SHC recipients and non-recipients using pooled logistic regression with adjustments for baseline confounders. Five sensitivity analyses are employed, including per-protocol analysis, changing eligibility criteria and negative outcome control analysis.

Ethics and dissemination

This protocol obtained the approval from Kyoto University Graduate School and Faculty of Medicine, Ethics Committee (R2448). The findings will be disseminated to academic conferences, and published in a peer-reviewed journal.

Strengths and limitations of this study.

• This study examines the effectiveness of Specific Health Check-ups (SHC) in Japan—a nationwide programme that was launched in 2008 for the primary prevention of obesity-related disease—using a large-scale healthcare database.

• We adopt the target trail emulation framework that applies principles of randomised trials to our observational study.

• By following the structured process of a target trial emulation framework, we expect that the limitations inherent to the observational studies will be minimised.

• Unmeasured confounding factors, such as the preference for healthy lifestyle or socioeconomic status, can be a potential limitation; we are planning a series of sensitivity analyses to assess the impact of such factors.

• Disease misclassification due to under-reporting can occur in a subset of the population; we expect that the long follow-up period of up to 10 years may minimise the influence of this bias.

Introduction

Obesity is the driver of several chronic illnesses, including type 2 diabetes, hypertension and dyslipidaemia, as well as cardiovascular disease as a consequence.^{1 2} The healthcare burden of obesity-related, non-communicable diseases not only represents a high level of attributed mortality and morbidity³ but also represents a significant economic impact.⁴ Given the rising trend, a strategy against obesity (or high body weight) is needed at both the individual and community levels, including the implementation and evaluation of evidence-based interventions.³

In 2008, Japan launched a programme known as Specific Health Check-ups (SHC), consisting of universal nationwide health screening followed by personal counselling for persons identified at risk of developing metabolic syndrome.^{5 6} All adults aged 40–74 years are required by law to participate every year, independent of the insurance type. Compared with other types of check-ups, this programme is unique in that its objective is the primary prevention of obesity-related illness.⁷ Insurers fully cover SHC costs without the personal fees needed. This programme aims to proactively manage the lifestyle of each person to prevent metabolic syndrome⁸—at-risk condition for future cardiovascular diseases. The SHC programme focuses on central obesity, characterised by the measurement of waist circumference.⁹ Despite the legal request, the SHC participation rate has been estimated at only 50%.¹⁰

There have been arguments against the health benefits of general health check-ups in recent years.^{11 12} These previous studies mostly examined the effectiveness of health check-ups in the secondary prevention of chronic illnesses, and the effectiveness in the primary prevention—the principal goal of the SHC programme—has not been investigated in the general population. Randomised controlled trials (RCTs) are the ideal solution to answer this open question, but they are not feasible for the programme already up and running throughout the nation. To investigate the effectiveness of SHC service on the primary prevention of obesity-related illnesses—namely, type 2 diabetes and hypertension—our planning study uses observational data combined with the research framework referred to as the target trial emulation.^13–15

Methods and analysis

Overview of the target trial emulation framework

Target trial emulation is the framework for causal inference using observational data. The procedure consists of two steps.¹⁶ First, investigators develop a protocol of ideal RCT—or the target trial—to answer the question of interest. Second, the protocol is modified to align with the structure of observational data by adhering the ideal protocol as much as possible. This process is empirically shown to prevent self-inflicted biases that are common in observational studies.¹⁷ Several emulated trials using observational data, such as administrative healthcare databases or disease registries, have been published in recent years.^18–20 There is some degree of methodological variation in trial emulation, as discussed later.

We first describe the study data source and then detail the key protocol components: eligibility criteria, treatment strategy, assignment procedure, follow-up period, outcome, causal contrast of interest and analysis plan.¹³

Data source

We use the healthcare database organised by a commercial data vendor (JMDC, Tokyo, Japan).^{21 22} The JMDC collects data from health insurance societies in Japan, consisting of ledgers of insured individuals, claims (for hospitalisation, outpatient treatment and drug prescription) and health check-up results, with information linked to personal identifiers. Data on claims are available from January 2005 onward, and health check-up results are available from April 2008 onward. As of the end of 2021, the cumulative number of enrollees is over 13 million from approximately 250 health insurance organisations. The diagnoses of each enrollee are maintained by International Statistical Classification of Diseases and Related Health Problems, 10th revision (ICD-10) codes, and prescription histories are in Anatomical Therapeutic Chemical (ATC) codes. Data are anonymised by JMDC.

Records of other types of legally required health check-ups (eg, check-ups for occupational health hazard monitoring in the workplace) are also part of the JMDC database, and there is no indication as to whether the records are SHC-derived. If waist circumference is included, we regard the data as an SHC record circumference because this is not mandatory for other types of check-ups.

The target trial

We first describe the protocol of an ideal randomised trial. We are now planning two separate studies with outcomes of the onset of type 2 diabetes and hypertension. For simplicity, we explain the diabetes study as an example here, but it is also applicable for hypertension with minor modifications.

Eligibility criteria

Persons aged 40–74 years who are metabolically healthy (ie, without hypertension and type 2 diabetes) at baseline and have never received SHC before are eligible for this study.

This study will not include persons who are unlikely to have general check-up services, such as severely disabled individuals or residents of long-term care facilities. We also exclude persons who have never received SHC but have received other check-up services because receiving other types of check-ups might influence the outcome of the check-up of interest via mechanisms such as lifestyle modification. In addition, patients with type 1 diabetes are excluded. As there are no contraindications for SHC, the exclusion criteria are kept minimal.

Treatment strategy

One or more receipt of annual SHC versus non-receipt. The items of the SHC programme are prescribed by the Ministry of Health, Labour and Welfare and include measurements for blood pressure and blood glucose level.²³ Usual care is given to both groups, including the measurement of blood pressure or blood glucose, if clinically indicated. Receiving SHC in subsequent years is permitted only in the intervention group.

Assignment procedure

Allocation is assigned by a randomised procedure into either the SHC receipt versus no SHC group and will be aware of the strategy they have been assigned to.

Follow-up period

For each individual, follow-up starts at randomisation and ends at 10 years after randomisation, the first diagnosis of type 2 diabetes, loss to follow-up, or March 2022, whichever occurs first.

Outcome

New diagnosis of type 2 diabetes. Data are scheduled to be collected at annual regular visits, and thus the unit of analysis is a 1-year interval.

Causal contrast of interest

The difference in 10-year incidence proportions of the outcome and the HR during the 10-year follow-up. For each measure, two causal contrasts are possible: intention-to-treat (ITT) basis and per-protocol-basis. The ITT effect compares the assignment of receipt versus non-receipt of SHC at baseline regardless of the subsequent use of annual SHC. The per-protocol effect compares the intended strategies under perfect adherence to assignment, assuming the absence of receiving SHC in the non-receipt group. The causal contrast is set to the ITT principle to maintain the comparability introduced via randomisation.²⁴

Analysis plan

The 10-year risks of incident diabetes among individuals assigned to each intervention strategy are estimated by the Kaplan-Meier method, assuming random censoring within each group. The pooled logistic regression model will be used to estimate the HR.

Emulation of the target trial

We next develop the study protocol using observational data (ie, the JMDC database) by synchronising it with that of the target trial as much as possible.

Eligibility criteria

The key inclusion and exclusion criteria are the same as those in the target trial protocol.

To ensure the prior history of (no) receipt of SHC and diabetes, a minimum of 12 months of JMDC database enrolment prior to cohort entry is required as the data lookback period. At least one visit to the medical institution, regardless of reason, is required within all lookback periods (alternatively within the fixed-period, such as 12 or 24 months) to ensure no prior hypertensive/diabetic history.

The history of type 2 diabetes is identified by the combination of the recorded diagnosis of diabetes (ICD-10 codes: E11 and E14) and the prescription of antidiabetic drugs (ATC10 code: A10) in the outpatient setting. This definition was reported as yielding high accuracy in the JMDC database.²⁵ Using the combination of recorded diagnosis and prescription records, the sensitivity and specificity were reportedly 78.6% and 99.6% for diabetes, and 74.5% and 98.2% for hypertension, respectively. Additionally, the positive predictive value and negative predictive value were 94.2% and 98.1% for diabetes, and 92.9% and 92.5% for hypertension, respectively.

Treatment strategy

Same as the target trial protocol.

Assignment procedure

Participants are classified into either the receipt or the non-receipt group according to the SHC receiving status at the baseline point, at which eligibility is assessed for each participant. Unlike RCTs, however, participant characteristics can differ between groups. To be comparable, the baseline characteristics will be adjusted for (see ‘analysis plan’).

We will employ multiple nested trials to emulate the target trial.^{13 26} In this approach, participants are repeatedly allowed to enter into the multiple trials that have the common protocol but start at distinct baseline points. If a person meets eligibility criteria multiple times, she or he sequentially enters the trials and is assigned to the groups according to the receipt or the non-receipt of SHC at that point. Finally, data from all trials are pooled into one data set; the within-person correlation is accounted for by statistical analysis (see ‘analysis plan’).

In this study, we will start the nested trials annually during the study period (ie, from 2008 to 2021); each participant candidate will be recruited to the distinct trials every year, assessed by her or his eligibility, assigned to the treatment group, and followed-up (figure 1).

Figure 1.

Illustrative scheme of the nested trials. FY, fiscal year; SHC, Specific Health Check-ups.

Trial #1 (fiscal year 2008)

We enrol the person who meets the eligibility criteria in 2008 (ie, the first SHC year).

In this population, no one has received SHC before; if an eligible person has ever received other types of check-ups, she or he will be removed from the cohort. Persons receiving SHC within this calendar year constitute the exposure group (regardless of whether the person will participate in the subsequent SHC screening), whereas persons without SHC service in the same period are classified into the control group; both groups are mutually exclusive.

Trial #2 (fiscal year 2009)

We enrol the person who meets the eligibility criteria in 2009; their inclusion criteria are: no prior history of SHC receipt and no diabetes/hypertension diagnosis, and: JMDC data history of ≥12 months, accompanied by at least one medical visit during this period. Persons receiving SHC within this calendar year constitute the exposure group, and persons without SHC service are classified into the control group, similar to Trial #1.

These sequential trials are repeated until we create a cohort of fiscal year (FY) 2021, and then they are compiled into a single cohort. Note that we can measure baseline confounders at the explicit time points in both the SHC recipient and non-recipient groups and that baseline characteristics differ even for the same person in the different trials (eg, age).

Follow-up period

All individuals will be followed from the beginning of each trial to the first of the following events: disenrolment from the JMDC database (including death), the outcome occurrence, 10 years since the treatment assignment or the administrative end of the study (expected in March 2022).

Outcome

The primary outcome is a new diagnosis of type 2 diabetes. This information is identified by the combination of the diagnosis of diabetes and the initiation of glucose-lowering agent(s) at the outpatient encounter; the same codes (ICD-10 codes of E11 and E14 for the diagnosis and ATC code of A10 for prescription) are used. The unit of analysis is a 1-year interval, as identical to that of the target trial.

Causal contrast of interest

It is unlikely that all persons fully adhere to the assigned intervention throughout the observation. The persons in the non-SHC group would violate the assignment if they receive the subsequent check-up, which can occur in a subset of population with some reasons (eg, change in health concern); please note that, by definition in our study protocol, the individuals in the SHC group are regarded as adhering to the intervention even when they do not receive thereafter check-up. As such, the per-protocol effect, in which every participant completely adheres to the intervention,²⁷ is difficult to interpret, and we thus prefer to estimate the ITT effect. However, we additionally seek to examine the per-protocol effect in the sensitivity analysis because the censoring pattern is typically assumed to be random in ITT analysis, but this may not be the case in SHC.

Analysis plan

We will first create a Kaplan-Meier plot adjusted by baseline covariates with the inverse probability weighting method,²⁸ in order to estimate the event rate in each group that accounts for the imbalance in baseline characteristics. The variables to calculate the probability of receiving SHC include age at cohort entry, sex, the number of medical visits (expressed in quantile or quartile depending on the distribution), hospitalisation within 12 months (binary), the medical history and comorbidity (expressed as binary variable for each disease category based on ICD codes) and employment status (binary).²⁹ Regarding the past illnesses and comorbidities, diabetes, hypertension, pregnancy-related disorders (ICD codes of O.xx) and illness specific to the neonatal period (ICD-10 codes of P.xx) will not be used for the adjustment, but all other comorbid condition swill. The presence of the medical histories will be screened using three-digit ICD-10 codes. The weights are estimated from a logistic regression model for predicting treatment assignment, and stabilised weights are finally used for adjustment. We will check the balance with standardised differences across each measured covariate before and after the weighting.

To estimate the HR (or its equivalence), the weighted Cox proportional hazard model or the pooled logistic regression^{30 31} will be used where appropriate. As mentioned earlier, censoring includes 10 years after study entry, withdrawal from the JMDC database, the incidence of type 2 diabetes or the administrative end of the study. A 95% CI is estimated with robust sandwich variance estimators or non-parametric bootstrapping.

It is uncertain whether our emulation framework is computationally efficient. The expected sample size is ~millions, and some participants will enter more than once. If we find that the calculation is computationally intensive, random sampling of the cohort (eg, 5–10%) would be the possible solution.³²

Finally, the protocol of the target trial and its emulation are summarised in table 1.

Table 1.

Summary of the target trial protocol and emulated protocol

Protocol components	Target trial (ideal RCT)	Emulation using observational data
Eligibility criteria	Persons aged 40–74 years.* No type 2 diabetes and hypertension. No history of SHC receipt and other types of health check-ups.	Similar to the target trial Data lookback period ≥12 months. ≥1 medical visit during the lookback period.
Treatment strategy	Receipt of SHC versus non-receipt	Same as the target trial
Assignment procedure	Randomisation	Randomisation is emulated by nested trial design
Follow-up period	10 years since randomisation	Disenrolment from JMDC database (including death). The administrative end of the study. 10 years after the enrolment in each emulated trial. (The outcome occurrence).
Outcome	New diagnosis of type 2 diabetes	Same as the target trial ICD-10 code: E11 and E14. ATC code: A10.
Causal contrast of interest	ITT principle Expected estimand: the effect of being assigned to each intervention, irrespective of any post-randomisation deviations (that can occur in both groups)	ITT principle* Expected estimand: the effect of being assigned to each intervention, irrespective of any post-randomisation deviations (that can occur in the non-SHC group only)
Analysis	Cox proportional hazard model	Pooled logistic regression

1: Modified in the sensitivity analyses.

*Modified in the sensitivity analyses.

ATC, Anatomical Therapeutic Chemical; ICD-10, International Statistical Classification of Diseases and Related Health Problems 10th revision; ITT, intention-to-treat; RCT, randomised controlled trial; SHC, Specific Health Check-ups.

Sensitivity analysis

Five types of sensitivity analyses are currently planned.

First, we will examine the per-protocol effect attributed to ‘non-adherence to the assigned intervention’. In addition to the censoring definition in the primary analysis, the receipt of SHC screening among the non-recipient controls in the subsequent years is also treated as censoring. For SHC recipients who do not have any further check-up service, we do not regard such an event as censoring because we are not sure how long the effect of SHC will be sustained. In this sensitivity analysis, because censoring (ie, the receipt of SHC among controls) can be non-random and thus lead to potential selection bias, we use the Inverse Probability Censoring Weights method.³³ The denominator of the inverse probability weight represents the probability that a participant adheres to her or his assigned group calculated via logistic regression, which is estimated using the updated information on the covariates (when applicable) used for baseline adjustment, years of follow-up and the number of SHC services (for non-recipients, this corresponds to zero).

Second, the inclusion criteria will change to persons between 40 and 69 years old. Because Japan has a special insurance plan for all elderly persons aged ≥75 years old, participant follow-up ends at the age of 74 years among JMDC enrollees. For this reason, persons who receive the SHC at the age of ≥70 years have a follow-up period of less than 5 years, which may be too short to observe the incidence of type 2 diabetes.

Third, we will relax the inclusion/exclusion criteria by enrolling persons who have coexisting diabetes or hypertension (eg, including persons with diabetes at baseline when the study outcome is set at hypertension) or persons who have received other types of check-ups. We expect that this modification will increase the generalisability.

Fourth, to ascertain the impact of selection bias and unmeasured confounding, we are planning two negative outcome control analyses³⁴ : gastric cancer and brain cancer. As the primary aim of the SHC programme is to reduce obesity-related disease, we expect that these two cancers will work as negative outcomes; we further discuss the rationale for these outcomes as negative outcomes in online supplemental file 1. These two cancer outcomes, however, have different features; the Japanese government prepares a gastric cancer screening programme for persons 40–74 years old but not for brain cancer. If the estimated risk of the two cancers differs, it can be explained by unmeasured factors such as preference for healthcare screening programme (although data on whether each person receives gastric cancer screening are not present in JMDC data set). This sensitivity analysis is limited to individuals without a history of index cancer.

Finally, if it is computationally feasible, we plan to conduct another sensitivity analysis to explore whether a different time scale (eg, monthly cohort entry rather than an annual basis) would yield similar or different estimate.

With respect to the second and third sensitivity analyses, the causal contrast of interest and analysis plan are the same as those in the primary analysis.

Patient and public involvement statement

The data source is anonymised, and patient and public involvement is not planned.

Ethics and dissemination

The original protocol of this paper obtained the approval from Kyoto University Graduate School and Faculty of Medicine, Ethics Committee (R2448). Because of the anonymised nature of the data, the consent of each JMDC enrollee is not required under the approval of the ethics committee. All data are fully anonymised before we obtain it from the data provider. Registration is not planned because the registration sites are typically prepared for RCTs. Findings will be disseminated to academic conferences, and published in the peer-reviewed journal(s).

Discussion

This study protocol describes the rationale and the study design of our research plan to examine the effectiveness of SHC in Japan using a large-scale healthcare database. By following the structured process of a target trial emulation framework, we expect that the limitations of observational studies will be minimised.

Bias and confounding are inherent limitations of observational studies.³⁵ In our case, a simple comparison of recipient versus non-recipient of SHC would introduce selection bias due to the different healthcare seeking behaviours. For example, persons who prefer a healthy lifestyle are more likely to receive SHC. In contrast, it is also possible that persons who are concerned about their health would access SHC more commonly because of some symptoms. Accordingly, the direction of bias arising from the different healthcare-seeking behaviours can be bidirectional and difficult to predict. One solution for such issues is to conduct an RCT. However, the SHC was adopted more than 10 years ago for every citizen aged 40–74. This situation does not allow us to conduct an RCT, but it also means that data has accumulated to evaluate the effectiveness of SHC with an observational study design. The goal of our research project is to analyse such observational data as analogous to randomised experiments as possible using the framework of target trial emulation.

The rationale and potential problem of SHC in Japan

The SHC programme primarily aims to prevent incident obesity-related diseases rather than to diagnose the unrecognised disease of each participant. This programme is delivered to all citizens aged 40–74 years, rather than focusing on the high-risk population. Some evidence has suggested the beneficial effect of SHC and personal counselling thereafter,^{5 36} although no conclusions have been reached.³⁷ However, it has also been pointed out that the volume and range of health check-up services in Japan are high, and whether additional check-ups may offer benefits for population health or the financing of the healthcare system remains unclear.⁶

Screening programmes may not work as intended. The illustrative example was the neuroblastoma screening programme for infants. Although several countries, including Japan, have adopted this screening programme to reduce neuroblastoma-related mortality, the screening is unlikely to reduce either the incidence of advanced disease or neuroblastoma mortality and is associated with overdiagnosis, leading to unnecessary treatment.³⁸ Similar to this issue, studies from other countries suggested that regular, population-based check-ups did not offer health benefits overall.¹¹ These lessons highlight the importance of the scientific basis of the population-level screening programme.

Comparison with other emulation approaches

The emulation approach involves variations and other potential schemes. Defining ‘time zero’ is crucial, particularly for the non-intervention group.¹³ Specifying ‘time zero’ helps minimise biases such as immortal time bias.³⁹ However, if participants have multiple eligible time points, choosing ‘time zero’ becomes complex. One approach is to use the beginning of eligibility of the programme (eg, the FY of 2008, or at the age of 40 years) as the study entry point, but it has trade-offs: evaluating the effectiveness of later screenings entry is not possible, and it may reduce statistical efficiency.³² Another alternative is to create two sets of clones for each individual, assign interventions and follow them.^{16 19} However, defining a uniform grace period, an essential component in the cloning approach,¹⁴ is challenging in the context of the programme due to the variability of check-up opportunities based on personal factors. Therefore, the cloning approach is not adopted.

Potential limitations

We hereby discuss the expected limitations. First, unmeasured confounding factors may remain. For example, body mass index, blood pressure level and its trajectory are prognostic factors of the subsequent onset of diabetes or hypertension. However, these factors are available only among the persons receiving SHC. Preference for a healthy lifestyle, including smoking status or alcohol consumption, is also an unmeasured factor that would influence non-communicable disease onset. Although socioeconomic status is also a potential unmeasured confounder, JMDC members are regarded as employees of large companies and their family members^{40 41}; thus, we assume that socioeconomic status is, to some extent, homogeneous in our study population. Second, the diagnosis of diabetes or hypertension may be under-reported, particularly in non-recipient populations, because these conditions are asymptomatic in the early phase of disease and are identifiable only by physical examination or laboratory testing. We expect that longitudinal data with up to 10 years of follow-up would minimise underdiagnosis, but it is difficult to predict in advance how long the data of each person will be available or how much this misclassification bias can be reduced.

Conclusion

We herein describe the study protocol, including the rationale and the study design of our research plan with a framework of target trial emulation. We hope that our future work will serve to improve population health and to spur policy implementation in Japan and other countries.

Data availability statement

Data may be obtained from a third party and are not publicly available. Data sharing is not allowed by the data provider.

Ethics statements

Patient consent for publication

Not applicable.

Ethics approval

Not applicable.