Evaluation of the impact of a recommended shift in glycaemic management targets for older adults on clinical outcomes and healthcare expenditures in Japan: a protocol for interrupted time-series analysis using a nationwide health insurance claims database

Abstract

Introduction

In 2017, the Japan Diabetes Society and Japan Geriatrics Society published the Clinical Practice Guidelines for the Treatment of Diabetes in older adults, marking a major shift in glycaemic management policy for older adults. The guidelines represented a transition from conventional, uniform targets, originally developed for the general adult population, to stratified glycaemic goals tailored to the complex care needs of older patients, including comorbidities and frailty. Although the 2017 guidelines aimed to promote individualised care and reduce adverse events such as severe hypoglycaemia, the real-world impact on patient outcomes, clinical practice and healthcare expenditures has not been evaluated at the national level.

Method and analysis

A population-based interrupted time-series analysis will be conducted using data from the National Database of Health Insurance Claims of Japan, which captures nearly all insured healthcare encounters nationwide. This study will include individuals aged ≥65 years with diabetes who received insurance healthcare services between April 2016 and June 2019. Outcomes will be evaluated across three domains: patient outcomes, clinical practice and healthcare expenditures. Specifically, these will include the incidence of severe hypoglycaemia, acute coronary syndrome, hyperglycaemic emergencies such as diabetic ketoacidosis and hyperosmolar hyperglycaemic state, number of antidiabetic prescriptions and total healthcare expenditures. Primary analyses will use generalised linear mixed-effects models assuming Poisson or negative binomial distributions with adjustments for facility-level heterogeneity. Stratified analyses will be performed according to comorbidity burden, frailty status and receipt of relevant healthcare services. Sensitivity analysis will assess the robustness of the results using an alternative definition of severe hypoglycaemia.

Ethics and dissemination

This study was approved by the Ethics Committee of the National Centre for Geriatrics and Gerontology (No. 1752), and the need for informed consent was waived owing to the use of anonymised administrative data. These findings will be disseminated through peer-reviewed publications and presentations at international academic conferences.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• This study uses the National Database of Health Insurance Claims of Japan (NDB), one of the largest healthcare databases in the world, to enable a comprehensive and population-wide evaluation of the impact of the 2017 clinical guidelines for diabetes care in older adults.

• The use of high-coverage administrative claims data allows for a near-complete ascertainment of insured healthcare utilisation, enhancing the generalisability of the findings to the older Japanese population.

• Subgroup analyses examine whether the guideline-associated changes in outcomes differ across clinical and implementation-related subgroups, including groups defined by uptake of nutritional and disease-management interventions (as proxies for individualised, multidisciplinary care).

• Nonetheless, the NDB lacks key clinical, socio-economic and provider-level variables (eg, cognitive and functional status and education and income), which may lead to residual confounding and limit assessment of individualised guideline adherence.

Introduction

Traditionally, clinical guidelines for diabetes care have been developed with a primary focus on the general adult population, often without fully accounting for age-related physiology, comorbidity or care need differences. Recently, however, in both research and clinical practice, older adults with diabetes have been increasingly recognised as a heterogeneous population that requires a more individualised treatment approach.¹ Several international guidelines, including those issued by the American Diabetes Association,² International Diabetes Federation³ and European Association for the Study of Diabetes,⁴ emphasise that overly strict glycaemic targets in older adults may increase the risk of harm without clear cardiovascular benefit. Accordingly, they have begun to incorporate more flexible targets and consistently endorse the principle of individualised treatment goals. Nonetheless, detailed frameworks for individualised glycaemic management, particularly those addressing functional status, cognitive impairment, polypharmacy and care dependency, remain limited in scope and implementation.

Against this backdrop, Japan has developed guidelines specifically targeting older adults with diabetes,^{5 6} representing a distinctive and internationally notable initiative.^{7 8} In response to Japan’s ageing population, the 2017 Clinical Guidelines for the Treatment of Diabetes in older adults fundamentally shifted the focus toward individualised haemoglobin A1c (HbA1c) targets tailored to functional and cognitive status, comorbidities and frailty, rather than applying uniform thresholds. The emphasis on individualised care represents the guideline’s core innovation, with the prevention of overtreatment and avoidance of severe hypoglycaemia positioned as important clinical consequences of this broader approach.^{9 10} These features reflect the response of Japan to its rapidly ageing population, in which the clinical and societal imperatives for personalised diabetes care are particularly pressing.

Although the guidelines have been widely disseminated and regarded as a turning point in geriatric diabetes management in Japan, their actual impact on clinical practice, patient outcomes and healthcare expenditures is yet to be assessed at the national level. Empirical evidence is warranted to determine whether the publication of the guidelines has led to measurable improvements in care quality and healthcare resource utilisation.

We hypothesise that the 2017 guidelines have led to a reduction in the incidence of severe hypoglycaemia and acute coronary syndrome (ACS) without increasing risk of hyperglycaemic emergencies, as well as changes in antidiabetic prescriptions and healthcare expenditures among older adults with diabetes in Japan.

Aim

This study aims to evaluate the impact of the revised glycaemic control targets for older adults, proposed in the Clinical Guidelines for the Treatment of Diabetes in older adults 2017, on patient outcomes, clinical practice and healthcare expenditures across Japan. By quantitatively assessing the guideline’s influence on these domains using nationwide data, we will provide robust evidence of its effectiveness. In addition, subgroup analyses will identify patient populations in whom the guideline’s impact is most pronounced. Together, these findings will not only demonstrate the real-world effectiveness of the guideline but also inform future revisions and dissemination strategies, thereby contributing to improved diabetes care and policy planning for older adults.

Method and analysis

Study design

This population-based observational study uses a nationwide administrative claims database from April 2016 to June 2019.

Study setting

Nationwide hospitals and outpatient clinics in Japan.

Data source

Data used in this study will be obtained from the National Database of Health Insurance Claims of Japan (NDB). Developed and maintained by the Ministry of Health, Labour and Welfare (MHLW), the NDB is a comprehensive administrative claims database that covers nearly the entire Japanese population, owing to the universal health insurance system of the country. All medical and pharmaceutical claims submitted electronically by insurers are collected by the MHLW and subsequently anonymised.

The MHLW provides NDB data to policymakers and researchers for specific purposes, such as policy planning or academic research, in accordance with predefined protocols. After approval by the MHLW review committee, the anonymised dataset was provided to the research team in April 2024; therefore, the study start date is April 2024, and the planned study end date (completion of analyses) is March 2026. All procedures will comply with the NDB usage guidelines and data-handling regulations. The data will be accessed only by designated researchers in a secure room at the National Centre for Geriatrics and Gerontology. Investigators have access to the anonymised, record-level claims data provided by the MHLW for the approved study purpose.

Participants

This study will include individuals aged ≥65 years with diabetes who used healthcare services covered by health insurance between April 2016 and June 2019. Eligible participants will be those diagnosed with diabetes (International Classification of Diseases, 10th Revision (ICD-10) codes E10–E14) who received at least one prescription for antidiabetic medication, including alpha-glucosidase inhibitors, sodium–glucose cotransporter-2 (SGLT2) inhibitors, thiazolidinediones, biguanides, dipeptidyl peptidase-4 (DPP-4) inhibitors, glucagon-like peptide-1 (GLP-1) receptor agonists, sulfonylureas, glinides, insulin preparations, or combination agents.¹¹ Diagnoses flagged as ‘suspected’ in claims records will be excluded. A participant flowchart is shown in figure 1.

Figure 1. Flowchart of participant inclusion. DPP-4, dipeptidyl peptidase-4; ICD-10, International Classification of Diseases, 10th Revision; SGLT2, sodium–glucose cotransporter-2.

Based on the 2017 Patient Survey (Government Statistics Code: 00450022) conducted by the MHLW, the potential target population is estimated at approximately 37 000 and 1 621 000 older adults with type 1 and 2 diabetes, respectively,

Intervention

The intervention of interest will be the publication of the guidelines, released in June 2017. Within the framework of the interrupted time-series (ITS) design, this publication will be treated as a time-point intervention to evaluate its potential nationwide impact.

These guidelines are considered major turning points in the quality of diabetes care for older adults in Japan, as they introduced stratified glycaemic targets based on age-related factors, such as physical and cognitive function and the level of independence in activities of daily living, marking a shift away from uniform treatment goals toward more practical and comprehensive recommendations tailored to the heterogeneous needs of older adults.

Measurement

The following monthly aggregated variables will be assessed at the facility level within the study population.

Outcomes: patient outcomes

Incidence of severe hypoglycaemia

Based on a previous study,¹² severe hypoglycaemia is the presence of a diagnosis code for hypoglycaemia (ICD-10: E15, E100, E110, E140, E160, E161 or E162) in conjunction with the administration of a 50% intravenous glucose injection on the same day (code: 620006649, 620002599, 640460006, 643230048, 643230050, 643230052, 640412069 and 640412070).

Incidence of acute coronary syndrome

With reference to a previous study,¹² ACS events, including acute myocardial infarction and unstable angina, will be defined as the occurrence of an emergency percutaneous coronary intervention, identified using procedure codes (150375210, 150375310, 150374910, 150375010 and 160107550).

Incidence of hyperglycaemic emergencies

The guidelines encourage a shift from uniform, strict glycaemic control towards individualised and potentially de-intensified management. While this approach aims to reduce overtreatment and hypoglycaemia, potential risks associated with treatment relaxation must also be considered. Previous studies have reported that hyperglycaemic emergencies such as diabetic ketoacidosis (DKA) and hyperosmolar hyperglycaemic state (HHS) occur more frequently in patients who are not taking antidiabetic medications, highlighting treatment non-adherence as a major risk factor.^{13 14} To address this concern, we will evaluate the incidence of DKA and HHS.

Based on prior research,¹⁵ DKA will be defined using ICD-10 diagnosis codes E10.1, E11.1, E12.1, E13.1 and E14.1. HHS will be defined using ICD-10 codes E10.0, E11.0, E12.0, E13.0 and E14.0. Because the HHS codes encompass both hyperglycaemic coma and hypoglycaemic coma, patients who received treatment for hypoglycaemia during the same encounter will be excluded to avoid misclassification.

Outcomes: clinical practice

Number of antidiabetic prescriptions

This outcome will be defined as the number of monthly antidiabetic medication prescriptions. Antidiabetic agents will include the following drug classes^{11 16}: alpha-glucosidase inhibitors, SGLT2 inhibitors, thiazolidinediones, biguanides, DPP-4 inhibitors, GLP-1 receptor agonists, sulfonylureas, glinides, insulin preparations and combination agents.

Outcomes: healthcare expenditures

Total healthcare expenditures

Monthly total healthcare expenditure will be calculated using health insurance claims data and expressed in Japanese yen. The costs will be assessed from the perspective of the payer.

Population characteristics

Age will be categorised into 65–69, 70–74, 75–79, 80–84, 85–89, 90–94, 95–99 and 100 years or older. Sex will be classified as male or female. The comorbidity burden will be assessed using the Charlson Comorbidity Index (CCI),¹⁷ including the average monthly CCI score and the distribution of the number of comorbid conditions per person. Frailty will be evaluated using the claims-based frailty index (CFI)¹⁸ and categorised into three levels: robust, pre-frail and frail. We will calculate the monthly prevalence of comorbidities identified in previous studies¹⁹,²¹ as risk factors for severe hypoglycaemia: dementia, renal disease, congestive heart failure, myocardial infarction, cerebrovascular disease, malignancy, chronic pulmonary disease, mild liver disease, rheumatologic disease and peripheral vascular disease, as well as fracture, depression, osteoporosis, knee or hip osteoarthritis and periodontal disease. Furthermore, the monthly prevalence of diabetes-related complications will be assessed, specifically diabetic retinopathy and neuropathy.²⁰ Finally, we will determine the monthly proportion of individuals who received nutritional or disease management interventions, which will be identified using relevant procedure codes and will include outpatient nutritional counselling, inpatient nutritional counselling, group-based counselling, diabetes dialysis prevention programmes, home-visit nutritional counselling, diabetes complication management and specified disease management services. The ICD-10 and procedure codes used to define these conditions and services are detailed in online supplemental material.

Statistical analysis

The study outcomes will include patient outcomes (incidence of severe hypoglycaemia, ACS, DKA and HHS), clinical practice indicators (number of antidiabetic prescriptions) and healthcare expenditures. All outcomes will be assessed monthly. All statistical analyses will be conducted using R version 4.3.1, and two-tailed p-values<0.05 were considered statistically significant.

Descriptive analysis

Each outcome variable as well as covariates, including age and sex distribution, CCI score, CCI condition number and frailty status, will be summarised on a monthly basis for the entire study population.

The frequency of events such as severe hypoglycaemia, ACS, DKA, HHS, antidiabetic prescriptions and other categorical variables will be presented as counts and proportions (n, %). Continuous variables will be summarised using means and SD or medians and IQRs, as appropriate.

Further, monthly trends of each outcome will be plotted to visually assess the changes before and after guideline publication. Additionally, monthly trends will be examined in key population characteristics, including age distribution, sex distribution, frailty level and comorbidity burden, to evaluate whether any substantial changes occurred in the study population composition between the pre-intervention and post-intervention periods.

Primary analysis

ITS analysis²² will be used to evaluate changes in outcomes before and after the publication of the guidelines and will be treated as the intervention point. The pre-intervention and post-intervention periods will be from April 2016 to June 2017 (15 months) and from July 2017 to June 2019 (24 months), respectively.

To estimate the level and trend changes associated with the intervention, generalised linear models with distributional assumptions and link functions selected according to the characteristics of each outcome will be used. Specifically, severe hypoglycaemia, ACS, DKA and HHS, as well as antidiabetic prescription numbers, will be modelled using a Poisson or negative binomial distribution with a log-link function. Conversely, total monthly healthcare expenditures will be modelled using a gamma distribution with a log-link function.

An offset term will be included in the model. To account for facility-level heterogeneity, mixed-effect models with random intercepts and slopes will be used for each medical institution. Seasonality will be adjusted using Fourier terms (sine and cosine functions).

Model diagnostics will include the examination of Pearson residuals, with autocorrelation assessed using Autocorrelation Function (ACF) and Partial ACF (PACF) plots. If substantial serial correlation is identified, an AR (1) error structure will be applied to account for temporal dependence in the residuals.

Moreover, the presence of autocorrelation, defined as the correlation of a variable with itself over successive time intervals, will be evaluated, which, if unaddressed, may bias the effect estimates in time-series analyses. Following Bernal et al,²² the autocorrelation function and partial autocorrelation function plots will be inspected and the Durbin–Watson test conducted. If residual autocorrelation remains after adjusting for seasonality, it will be corrected using an autoregressive integrated moving average–type error structure or Newey–West standard errors.

The basic model structure is as follows:

$${\displaystyle {\displaystyle \begin{array}{l}log\left(E\left[Y_{it}\right]\right)=\beta 0+\beta 1\cdot tim{e}_{it}+\beta 2\cdot treatmen{t}_{it}+\beta 3\cdot \left(tim{e}_{it}\times treatmen{t}_{it}\right)\\ +{\mathrm{u}}_i+\mathrm{l}\mathrm{o}\mathrm{g}\left(denominato{r}_{it}\right)+{ϵ}_{\mathrm{i}\mathrm{t}}\end{array}}}$$

Y_it：number of severe hypoglycaemia events in facility i, during month t

time_it：time since the start of the observation period (April 2016), treated as a continuous variable ranging from 1 to 39

treatment_it：intervention indicator (coded as 1 for July 2017 onward and 0 otherwise)

time_it × treatment_it：post-intervention slope (trend) change

u_i：facility-level random intercept

denominator_it：number of unique patients with diabetes who received care at facility i during month t (‘patient-months’), used as an offset term to model event rates

ε_it：error term

Secondary analysis

Subgroups will be defined based on four prespecified indicators derived from population characteristics, selected based on previous research and clinical relevance, to explore whether the effect of guideline revision varies across these subpopulations. Notably, severe hypoglycaemia is associated with multiple clinical risk factors, including multimorbidity, dementia and renal disease.¹⁹,²¹²³ The glycaemic targets proposed in the 2017 diabetes treatment guidelines for older adults were designed to accommodate such diverse clinical profiles—particularly multimorbidity and frailty—in this population, and their impact may be more pronounced in high-risk groups.

First, subgroups will be defined according to the number of comorbid conditions, as specified in the CCI,¹⁷ categorised as 0, 1–2, 3–4 or ≥5 comorbidities. Additional subgroups will be defined based on the presence of specific comorbidities and diabetes-related complications identified as risk factors for severe hypoglycaemia in previous studies¹⁹,²¹ and by expert clinical judgement, including dementia, renal disease, congestive heart failure, myocardial infarction, cerebrovascular disease, malignancy, chronic pulmonary disease, mild liver disease, rheumatic disease and peripheral vascular diseases.

Second, subgroups will be defined according to the frailty status, assessed using the CFI,¹⁸ and categorised as robust, prefrail or frail.

Third, subgroups will be defined based on claims indicating the provision of healthcare services, such as nutritional counselling or disease management interventions. Evaluating outcomes within these subgroups will clarify whether integrating supportive care services with individualised glycaemic targets is associated with greater improvements in clinical outcomes, thereby offering a novel perspective on the real-world impact of guidelines.

Finally, the effects of guidelines may depend on the degree of adherence of the provider. Subgroups will be defined by stratifying the 47 Japanese prefectures (primary administrative divisions) according to the density of board-certified diabetologists per 100 000 people, as cardiologists adhere to guidelines more consistently than internists or family/general practitioners do.²⁴

To evaluate the impact of the guidelines across subgroups, stratified ITS analyses²⁵ will be performed according to the following steps.

For each subgroup, the ITS model developed for the primary analysis will be applied. Before model fitting, monthly changes in the relevant population characteristics within each stratum will be examined, including group size, age and sex distribution, frailty level and comorbidity burden. Autocorrelation will also be assessed within each subgroup, using the same procedure as that in the main analysis.

From each stratified ITS model, the coefficients representing the level (β₂) and the trend (β₃) changes associated with the intervention will be estimated, which will be used to calculate incidence rate ratios (IRRs) as IRR=exp(β₂) and IRR=exp(β₃), respectively. Additionally, 95% CI for the IRRs will be computed using Huber–White robust standard errors or, if necessary, Newey–West heteroskedasticity and autocorrelation-consistent standard errors.

Finally, time-series plots will be generated for each subgroup, displaying predicted values and 95% CI, as well as caterpillar plots of IRRs and their CIs. These visualisations will be used to assess potential differences in the impact of guideline revisions across the subgroups.

Sensitivity analysis

As a sensitivity analysis, we will (1) redefine the outcome of severe hypoglycaemia by expanding the criteria for intravenous glucose administration to include 20%–50% glucose injections and (2) introduce different lag periods for the intervention effect (eg, lag 1, 3 and 6 months) into the model to account for the possibility that the impact of the guideline publication may require time to manifest. These alternative specifications will be used to assess the robustness of our findings regarding changes in severe hypoglycaemia incidence in relation to guideline publications.

Ethics and dissemination

The study was approved by the Ethics Committee of the National Centre for Geriatrics and Gerontology (No. 1752), and the need for informed consent was waived because of the use of anonymised data that did not include personal information. The findings of this study will be published in peer-reviewed scientific journals and presented at conferences.

The results of this study will be reported in accordance with the STROBE statement and the RECORD guideline to ensure complete reporting of the study design, data sources and analyses.^{26 27}

Patient and public involvement

There will be no direct patient involvement in this study.

Discussion

This study aims to evaluate the impact of the publication of the 2017 Clinical Guidelines for Diabetes Care in Older Adults using data from the NDB. The NDB contains nearly all electronic claims submitted under the Universal Health Insurance system of Japan, excluding those covered by public assistance.¹¹ As of 2017, over 98% of claims were submitted electronically.²⁸ This comprehensive coverage allows for a near-complete ascertainment of insured healthcare utilisation, thereby enabling a robust evaluation of the nationwide impact of guideline publications, which is the key strength of this study.

Nevertheless, administrative claims data have some limitations. The NDB lacks information on important socioeconomic and demographic factors, such as place of residence, educational attainment and income, which have been associated with potential overtreatment and risk of severe hypoglycaemia in previous studies.²⁹,³²

Moreover, in Japan, diabetes care is delivered not only by specialists but also by many general clinicians, which may lead to variation in how guideline recommendations are applied in routine practice. However, physician-level characteristics, including medical specialty or familiarity with the guidelines, are not captured in the dataset; accordingly, guideline uptake and adherence cannot be assessed directly. Consistent with this concern, a Japanese outpatient study reported limited agreement between measured HbA1c and guideline-based individualised target HbA1c, suggesting potential gaps between recommended individualised targets and real-world glycaemic management.³³

Despite these limitations, assessing the extent to which nationally disseminated clinical guidelines influence healthcare practices and patient outcomes at a population level remains a critical endeavour. If guideline publication does not substantially reduce severe hypoglycaemia incidence or health expenditures, a gap in dissemination or implementation may exist, suggesting the need for additional policy efforts, such as provider education, performance-based incentives, or integration of individualised targets into routine care protocols.

Furthermore, planned subgroup analyses, particularly those based on frailty status and comorbidity burden, will provide important insights into the patient populations that are most responsive to individualised glycaemic targets. If significant effects are observed only among high-risk groups, these findings could help prioritise targeted interventions for frail or multimorbid older adults, enabling more efficient and equitable allocation of healthcare resources. Additionally, subgroup analyses will consider the provisional status of nutritional and disease management services. Although the 2017 guidelines did not explicitly prioritise nutritional therapy, their emphasis on individualised care for older adults with complex needs conceptually aligned with the integration of such services. If significant effects are observed among patients receiving these services, this would indicate a shift from uniform glycaemic targets toward a more comprehensive and patient-centred approach to diabetes management, representing a novel perspective for evaluating the impact of the guidelines.

Moreover, Japan represents a unique setting to study individualised and de-intensified diabetes management. As one of the most rapidly ageing societies worldwide, with a universal health insurance system and access to the NDB—one of the largest comprehensive claims databases globally—Japan is uniquely positioned to generate evidence that is not only relevant domestically but also highly generalisable to other countries facing similar demographic transitions. Thus, the findings from this study may have broad international implications for advancing personalised diabetes care in ageing populations.