Association of sustained integrative-care use with mortality and immobility-related complications among postfracture older adults in South Korean long-term care hospitals: a population-based, propensity score-matched, retrospective cohort study

Abstract

Objectives

To investigate whether sustained integrative care, combining traditional Korean medicine (TKM) with conventional medical care, is associated with lower mortality and reduced risk of major medical complications among older adults undergoing postfracture rehabilitation in long-term care hospitals (LTCHs).

Design

A nationwide, population-based, retrospective propensity score-matched cohort study.

Setting

LTCHs in South Korea, using the National Health Insurance Service customised claims database (2014–2023).

Participants

Older adults (aged ≥65 years) newly admitted to LTCHs with a fracture diagnosis were included in the cohort. Patients were matched 1:1 using propensity scores to an integrative care group (receiving acupuncture, moxibustion or cupping alongside conventional medical care) and a usual care group (receiving conventional care alone), resulting in a matched cohort of 40 720 participants (20 360 per group) with a mean follow-up of 3.0 years.

Primary and secondary outcome measures

The primary outcome was all-cause mortality. Secondary outcomes included the incidence of major immobility-related complications: pneumonia, deep vein thrombosis, pressure ulcers and urinary tract infections. The propensity score-matched groups were compared with multivariable Cox proportional hazards regression models.

Results

Compared with the usual care group, the integrative care group exhibited significantly lower risks of all-cause mortality (adjusted HR (aHR) 0.79 (95% CI 0.77 to 0.81)) and major complications, including pneumonia (aHR 0.72 (95% CI 0.69 to 0.76)), deep vein thrombosis (aHR 0.71 (95% CI 0.58 to 0.88)), pressure ulcers (aHR 0.76 (95% CI 0.72 to 0.80)) and urinary tract infections (aHR 0.75 (95% CI 0.70 to 0.80)).

Conclusions

Integrative care during postfracture rehabilitation in LTCHs is associated with lower mortality and fewer immobility-related complications in older adults. These findings support integrative rehabilitation as a potential strategy for postfracture care, although confirmation in prospective studies is warranted.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• This study used a national public insurance claims database linked to the death registry, allowing for high representativeness and reliable measurement of mortality and various medical complications.

• Propensity score matching and sequentially adjusted Cox proportional hazards models were applied to improve comparability between the study groups with respect to measured covariates.

• The study was limited by potential residual confounding from unmeasured clinical variables, such as exact fracture severity or specific functional metrics, that were unavailable in the administrative data.

• Due to administrative data constraints, there were insufficient details on individual treatment modalities and non-reimbursed medical services, limiting the ability to assess modality-specific effects.

• Because the database captured only long-term care hospital claims, baseline comorbidity burden was likely underascertained, potentially leaving residual confounding by comorbidity.

Introduction

Fractures in older adults represent a major global public health challenge, associated with substantial mortality, disability and long-term care dependence due to diminished physiological reserve and multiple comorbidities.^{1 2} For example, hip fractures are linked to an 8.4%–36% excess mortality within the first year.¹ Consistent with this substantial long-term burden, a recent Swedish cohort study of women aged 70 years or older reported a 1-year mortality rate of 27% after hip fracture and a mean survival of only 2.9 years following the first hip fracture.³ Among the various fragility fractures, only 52.3% of previously independent patients regain their functional status 1-year postinjury.² Driven by rapid population ageing, this clinical burden has escalated markedly in South Korea. Hip fractures occur at a rate of 181.5 per 100 000 individuals aged 50 years or older, with a 1-year mortality rate of up to 17.2%.⁴ Concurrently, the annual incidence of distal radius and vertebral fractures has reached 474.1 and 556.38 per 100 000, respectively, in older people.^{5 6} These figures emphasise the severe impact of fractures on the geriatric population and the need to optimise postfracture care strategies.

South Korea operates a unique dual structure for long-term care, which comprises residential nursing homes and medical long-term care hospitals (LTCHs). Unlike nursing homes, LTCHs are specialised medical institutions designed to provide subacute care and intensive rehabilitation. A distinguishing feature of these facilities is the colocation of both conventional medical doctors and traditional Korean medicine (TKM) doctors, facilitating a collaborative or integrative care model.^{7 8} Reflecting the growing demand for complex medical management, the number of frail older adults admitted to LTCHs has surged in recent years.⁹ Within this population, severe functional decline following a fracture is a primary catalyst for LTCH admission.¹⁰ Consequently, many older patients recovering from fractures in LTCHs receive national insurance-covered TKM interventions, such as acupuncture, moxibustion and cupping therapy, alongside standard medical care.⁸

Prior observational studies suggest that traditional medicine may improve clinical outcomes following a fracture. For instance, a nationwide retrospective cohort study in Taiwan reported that receiving acupuncture after hip fracture surgery reduced the risks of mortality and readmission by 59% and 36%, respectively.¹¹ Similarly, Chinese herbal medicine has been shown to decrease the risk of reoperation by 36% in patients with hip fractures and lower the risks of subsequent fractures (44%) and long-term mortality (31%) in those with osteoporosis.^{12 13} However, this existing evidence largely focuses on single fracture types or general adult cohorts rather than highly vulnerable geriatric patients. Consequently, robust evidence on the long-term impact of integrative medicine, specifically among older adults with fractures within the Korean LTCH setting, is critically lacking.

To address this critical knowledge gap, we conducted a nationwide retrospective cohort study using a customised claims database from the South Korean National Health Insurance Service (NHIS). The primary objective of this study was to evaluate the association between sustained integrative care, defined as the combination of TKM and conventional medical care, and mortality among older adults newly admitted to LTCHs with fractures, compared with those receiving usual care alone. Furthermore, we aimed to investigate the impact of this integrative approach on the incidence of major clinical complications, including pneumonia, deep vein thrombosis (DVT), pressure ulcers and urinary tract infections (UTIs). Ultimately, this study seeks to provide robust, real-world evidence to guide optimal, integrative postfracture care strategies for the geriatric population.

Methods

This study is reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines.¹⁴

Study design and data source

This was a nationwide, population-based retrospective cohort study using the NHIS administrative claims database in South Korea. As the sole public insurer, the NHIS covers nearly the entire population, approximately 97% through the National Health Insurance (NHI) programme and the remaining 3% through the Medical Aid programme.^{15 16} The NHIS database provides comprehensive longitudinal data on reimbursed healthcare utilisation with near-complete population coverage, encompassing demographic characteristics, healthcare provider information, diagnostic codes (based on the International Classification of Diseases, Tenth Revision (ICD-10)), procedures, prescriptions and exact dates of death, which are linked with the national death registry from Statistics Korea.¹⁶ Because the conventional medical care and TKM treatments provided in LTCHs are mostly reimbursable, the care service for this study is largely captured. However, as the database comprises administrative claims compiled for reimbursement, non-reimbursed services are not captured.

The overall study period spanned from 1 January 2014 to 31 December 2023, the most up-to-date period available when the data were obtained from the NHIS. Within this timeframe, the specific cohort enrolment period was established from 1 January 2015 to 31 December 2021, to incorporate a minimum 1-year washout period and a minimum follow-up duration of 2 years. The index date was defined as the start date of the first LTCH admission for the target fracture diagnosis.

Study population and eligibility criteria

We included patients aged 65 years or older at the time of their first admission to an LTCH who had a primary or secondary diagnosis of fracture on inpatient claims during the enrolment period (2015–2021). The target fractures encompassed vertebral (S220, S229, S320, S329, M484, M485 and T08), femur (S720, S721 and S729), shoulder (S422, S423 and S429), wrist (S525, S526 and S529), and other osteoporotic or unspecified fractures (M80 and T142).^{17 18} The femur fracture category included proximal femur/hip fractures (S720 and S721) as well as fractures of an unspecified part of the femur (S729), and therefore was not restricted exclusively to proximal femur/hip fractures. We opted to include a comprehensive spectrum of fractures to reflect the real-world clinical burden and demographics of older adults requiring postfracture rehabilitation in LTCHs.

To minimise potential biases, we sequentially applied exclusion criteria. First, patients were excluded if they had any LTCH inpatient claim during the washout period (365 days prior to the index date) to limit the cohort to newly admitted patients. Second, we excluded those with a diagnosis of S02 (fracture of skull and facial bones) or S12 (fracture of neck) on the index date to remove traumatic injuries (eg, traffic accidents), thereby focusing on the fragility fractures. Third, patients who died during the 30-day landmark period after the index date were excluded to minimise immortal time bias. Fourth, we excluded patients with a length of stay of fewer than 30 days across all admission episodes (allowing a 14-day gap between consecutive care episodes). Because LTCHs are designed for subacute care and long-term rehabilitation, it was assumed that stays of less than a month do not reflect the clinical intent of these facilities, as very short stays and early deaths are often related to terminal rather than rehabilitative admissions. This restriction ensures that patients have sufficient time for adequate treatment exposure. Fifth, we excluded patients who failed to meet a minimum frequency threshold for TKM interventions (defined as >0 but <1 session per 28 days). Given the nature of LTCHs, this criterion was adopted to exclude incidental treatments and ensure that the integrative care group (ICG) comprised patients receiving clinically meaningful, sustained TKM interventions. Finally, individuals with missing or invalid data were excluded.

Definition of study groups

In this study, conventional medical care refers to the full range of care routinely provided in LTCHs under the direction of medical doctors, including pharmacological treatment, osteoporosis management, rehabilitation therapy, nursing care and nutritional support, depending on individual patient needs and institutional practice. TKM treatments were identified from procedure claims and comprised acupuncture (including electroacupuncture), moxibustion and cupping therapy. Because these modalities are rarely delivered in isolation in LTCH practice, exposure was defined as receipt of one or more of them.

Based on the proportion of days covered (PDC), eligible patients were classified into two study groups. Patients were assigned to the ICG if they received sustained TKM treatments alongside conventional care, defined as a minimum treatment frequency of at least once per 28 days over the observed hospitalisation period (PDC ≥1/28). This specific threshold was set to align with the standard monthly clinical assessment and billing cycles characteristic of Korean LTCHs. In the context of long-term rehabilitation, this criterion ensures that the ICG represents patients receiving sustained maintenance-level integrative care rather than incidental one-off treatments. Conversely, patients who did not receive any TKM treatments during their hospitalisation (PDC=0) were classified into the usual care group (UCG), representing those receiving conventional care alone.

Propensity score matching

We employed propensity score matching (PSM) to minimise potential confounding by indication and to ensure baseline comparability between the two groups.¹⁹ Patients in the ICG were matched 1:1 to those in the UCG using nearest-neighbour matching without replacement, applying a calliper of 0.2.²⁰ Propensity scores were estimated using a multivariable logistic regression model that incorporated a set of baseline covariates, including demographic (sex, age) and socioeconomic factors (place of residence, income level), registered disability status, comorbidity burden (measured by the Charlson Comorbidity Index (CCI) score during the 365 days prior to the index date), and facility-level characteristics (LTCH establishment type and the number of beds). Between-group balance before and after PSM was assessed using the absolute standardised mean difference (SMD), with an SMD <0.1 indicating adequate balance.

Study outcomes

The primary outcome was all-cause mortality after the index date. Mortality status was ascertained using exact dates of death obtained through linkage to the Statistics Korea national death registry. This national-level linkage guarantees the comprehensive capture of both in-hospital and out-of-hospital deaths, thereby minimising follow-up loss. Follow-up for all-cause mortality continued from the index date until death or the end of the study period (31 December 2023), whichever occurred first.

The secondary outcomes were major medical complications frequently associated with geriatric fractures, specifically pneumonia (J12–J18 and J69), DVT (I80 and I82), pressure ulcers (L89) and UTIs (N39.0).^{21 22} These specific conditions were selected because older adults with fractures often experience prolonged immobility, severe functional decline and intense physiological stress, which substantially elevate their vulnerability to these adverse events. Postfracture pneumonia and UTIs, for instance, are highly prevalent and strongly linked to increased readmission rates, delayed functional recovery and long-term mortality.^{10 23–25} Furthermore, immobilisation and vascular injuries directly increase the risk of DVT and pressure ulcers, both of which severely compromise postfracture rehabilitation and quality of life.^{26 27} Each complication was ascertained from primary or secondary diagnoses, with a single qualifying claim. Participants were followed from the index admission until the first occurrence of the respective complication recorded more than 30 days after the index admission, death or the end of the study period, whichever occurred first.

Statistical analysis

Baseline characteristics of the matched cohort were summarised as means±SDs for continuous variables and as frequencies (percentages) for categorical variables. Formal normality testing was not performed, as such tests are overly sensitive to trivial deviations in large samples and parametric summaries remain valid under the central limit theorem.^{28 29}

To evaluate the association between integrative care and clinical outcomes, multivariable Cox proportional hazards regression models were used to estimate HRs and 95% CIs. We constructed sequentially adjusted models to control for potential residual confounders rigorously. Guided by the Andersen Behavioural Model of Health Services Use,³⁰ covariates were systematically adjusted: the crude model included only the exposure variable; Model 1 adjusted for predisposing factors (sex and age at admission); Model 2 further adjusted for enabling factors (place of residence, income level, LTCH establishment type and facility-level resources including the number of physicians, TKM doctors, registered nurses, nursing assistants and total beds); and Model 3 (the fully adjusted model) additionally included need factors (registered disability status and the CCI score). All adjusted covariates were measured at or before the index date, whereas postindex variables that could act as mediators were not included, so that the models estimated the overall association between integrative care and each outcome. Formal testing of the proportional hazards assumption was not performed, as such testing is not universally required and the HR can be interpreted as an average effect over the follow-up period.³¹

The overall comorbidity burden was calculated from diagnoses recorded over the 365 days prior to the index date using ICD-10 codes.³² The numbers of events and person-time at risk were estimated by group, and incidence rates were calculated as the number of events divided by the total person-time and expressed per 100 person-years. A complete-case approach was used, and participants with missing or invalid covariate data for propensity-score matching were excluded without imputation. Statistical significance was defined as a two-sided p value <0.05, with significance inferred when the 95% CI did not include 1.0. Statistical analyses were performed using SAS Enterprise Guide V.7.1 (SAS Institute, Cary, NC, USA).

Sensitivity analyses

To assess the robustness of our findings, several prespecified subgroup analyses were performed as part of the sensitivity analyses. We stratified the study population according to clinically relevant subgroups: sex (male or female), age group (65–74, 75–84 or ≥85 years), place of residence (capital area, non-capital metropolitan city or other), insurance type (Medical Aid or NHI), registered disability status (yes or no), CCI score (0, 1, 2–3 or ≥4) and fracture subtype (femur, shoulder, vertebral, wrist or unspecified). The fracture subtypes were not mutually exclusive, as patients could have more than one fracture type. For these analyses, separate PSM was performed within each subgroup to ensure baseline balance before conducting survival analyses. Following PSM, the identical sequential modelling approach (from the crude model to Model 3) was applied to each subgroup. Finally, forest plots visualising the treatment effects derived specifically from the fully adjusted models were generated using R software V.4.5.0 (R Foundation for Statistical Computing, Vienna, Austria). No formal interaction tests were performed. Therefore, the subgroup analyses were considered exploratory, and differences in statistical significance or point estimates across strata were not interpreted as evidence of effect modification. No adjustment was made for multiple comparisons in the subgroup analyses.

For the fracture-subtype subgroups, we additionally computed absolute event counts, person-years and incidence rates within each subtype. To further assess the robustness of the exposure definition, the associations for the primary and secondary outcomes were additionally estimated across mutually exclusive categories of increasing treatment frequency (1/28≤PDC<1/21, 1/21≤PDC<1/14 and PDC≥1/14) using the fully adjusted Cox proportional hazards model (Model 3), with a test for linear trend across the ordered categories, obtained by modelling treatment frequency as a single ordinal term. Because this test assumes a monotonic dose–response relationship, it was interpreted with caution.

To assess the robustness of the observed associations to potential unmeasured confounding, we calculated E-values for the fully adjusted point estimates and for the confidence limits closest to the null. The E-value represents the minimum strength of association, on the risk-ratio scale, that an unmeasured confounder would need to have with both integrative care and the outcome, beyond the measured covariates, to fully explain away the observed association. Because all-cause mortality was a common outcome, its E-value was additionally calculated using an approximation that accounts for the outcome frequency.³³

Patient and public involvement

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

Results

Cohort construction and baseline characteristics

A total of 231 702 patients aged ≥65 years with a fracture diagnosis were initially identified in the NHIS database between 2015 and 2021. After applying a 1-year washout period, 48 673 patients with previous LTCH admissions were excluded. We further excluded patients with skull/neck fractures (n=837), early mortality within 30 days (n=8210), length of stay <30 days (n=28 318), sporadic TKM use (n=71 319) and missing data (n=531). This sequential exclusion yielded an unmatched cohort of 73 814 patients, with 20 361 in the ICG and 53 453 in the UCG. After 1:1 PSM, the final analytic cohort comprised 40 720 patients (20 360 pairs) (figure 1).

Figure 1. Flow diagram of the study population selection process. LTCHs, long-term care hospitals; TKM, traditional Korean medicine.

The baseline characteristics of the study population before and after PSM are detailed in table 1. Before matching, patients in the ICG were younger (82.10±6.73 vs 83.01±6.96 years) and predominantly female (83.1% vs 76.8%) compared with those in the UCG. Following PSM, baseline covariates were exceptionally well balanced between the groups, with all SMDs well below the 0.1 threshold (online supplemental figure 1). In the matched cohort, the mean age was 82.11±6.73 years in the ICG and 82.19±6.96 years in the UCG, with comparable proportions of females (83.1% vs 82.7%). Furthermore, the finalised analytic cohort was characterised by a predominantly non-disabled status (∼74%) and admissions primarily to private or institutional LTCHs. The baseline comorbidity burden of the study population was generally low (mean CCI score: 0.18±0.77 vs 0.19±0.77), which may have been underestimated because the database did not fully capture diagnostic records from other medical institutions.

Table 1. Baseline characteristics of the study population before and after propensity score matching.

Variables	Unmatched			Matched
	ICG	UCG	SMD	ICG	UCG	SMD
	(n=20 361)	(n=53 453)		(n=20 360)	(n=20 360)
Sex
Male	3442 (16.9)	12 411 (23.2)	0.158	3442 (16.9)	3527 (17.3)	0.011
Female	16 919 (83.1)	41 042 (76.8)		16 918 (83.1)	16 833 (82.7)
Age	82.10±6.73	83.01±6.96	0.133	82.11±6.73	82.19±6.96	0.012
Place of residence
Capital area	7222 (35.5)	22 370 (41.8)	0.131	7222 (35.5)	7339 (36.0)	0.014
Non-capital metro city	4458 (21.9)	10 418 (19.5)		4458 (21.9)	4477 (22.0)
Other	8681 (42.6)	20 665 (38.7)		8680 (42.6)	8544 (42.0)
Income level
Very low	2913 (14.3)	6526 (12.2)	0.073	2913 (14.3)	2906 (14.3)	0.010
Low	3420 (16.8)	8843 (16.5)		3420 (16.8)	3441 (16.9)
Low-high	2668 (13.1)	6758 (12.6)		2667 (13.1)	2720 (13.4)
High-low	3570 (17.5)	9459 (17.7)		3570 (17.5)	3584 (17.6)
High	7790 (38.3)	21 867 (40.9)		7790 (38.3)	7709 (37.9)
Registered disability
None	15 048 (73.9)	39 831 (74.5)	0.082	15 048 (73.9)	15 068 (74.0)	0.015
Mild	3876 (19.0)	8944 (16.7)		3875 (19.0)	3793 (18.6)
Severe	1437 (7.1)	4678 (8.8)		1437 (7.1)	1499 (7.4)
CCI score	0.18±0.77	0.25±0.94	0.081	0.18±0.77	0.19±0.77	0.011
LTCH establishment type
Public	441 (2.2)	3024 (5.7)	0.182	441 (2.2)	438 (2.2)	0.003
Institutional	8034 (39.5)	20 747 (38.8)		8034 (39.5)	8009 (39.3)
Private	11 886 (58.4)	29 682 (55.5)		11 885 (58.4)	11 913 (58.5)
Number of LTCH beds	214.81±114.68	208.27±105.71	0.059	214.78±114.61	213.44±110.23	0.012

Data are presented as n (%) for categorical variables and mean±SD for continuous variables. Age, place of residence, income level, registered disability, LTCH establishment type and number of beds were assessed at the index date. Income level was categorised based on the type and premium of the National Health Insurance. The CCI score was calculated based on claims data from the 365 days preceding the index date.

CCI, Charlson Comorbidity Index; ICG, integrative care group; LTCH, long-term care hospital; SMD, standardised mean difference; UCG, usual care group.

Impact of integrative care on all-cause mortality

During a mean follow-up of 3.0 years (124 026 person-years), all-cause mortality occurred in 11 387 patients in the ICG and 12 939 patients in the UCG. In the crude analysis, integrative care was associated with a significantly lower risk of mortality (HR 0.81 (95% CI 0.79 to 0.83), p<0.001). This association remained highly stable throughout the sequential adjustment process for demographic and facility-level covariates (Models 1 and 2). Ultimately, in the fully adjusted model (Model 3) incorporating predisposing, enabling and need factors, the ICG demonstrated a 21% reduced risk of all-cause mortality compared with the UCG (adjusted HR (aHR) 0.79 (95% CI 0.77 to 0.81), p<0.001) (table 2). In the matched cohort, the incidence of all-cause mortality was 17.3 per 100 person-years in the ICG and 22.2 per 100 person-years in the UCG (online supplemental table 1).

Table 2. HRs for primary and secondary outcomes in the propensity score-matched cohort (n=20 360 pairs).

Outcomes	Number of events	Crude model		Adjusted model 1		Adjusted model 2		Adjusted model 3
	(ICG/UCG)	HR (95% CI)	P value	HR (95% CI)	P value	HR (95% CI)	P value	HR (95% CI)	P value
Primary outcome
All-cause mortality	11 387/12 939	0.81 (0.79 to 0.83)	<0.001***	0.83 (0.80 to 0.85)	<0.001***	0.79 (0.77 to 0.81)	<0.001***	0.79 (0.77 to 0.81)	<0.001***
Secondary outcomes
Pneumonia	2920/4439	0.70 (0.66 to 0.73)	<0.001***	0.72 (0.69 to 0.76)	<0.001***	0.72 (0.68 to 0.75)	<0.001***	0.72 (0.69 to 0.76)	<0.001***
Deep vein thrombosis	158/255	0.67 (0.55 to 0.82)	<0.001***	0.68 (0.55 to 0.82)	<0.001***	0.71 (0.57 to 0.87)	0.001**	0.71 (0.58 to 0.88)	0.001**
Pressure ulcer	2540/3616	0.74 (0.70 to 0.77)	<0.001***	0.75 (0.71 to 0.79)	<0.001***	0.75 (0.71 to 0.79)	<0.001***	0.76 (0.72 to 0.80)	<0.001***
Urinary tract infection	1587/2312	0.74 (0.69 to 0.79)	<0.001***	0.75 (0.70 to 0.80)	<0.001***	0.75 (0.70 to 0.80)	<0.001***	0.75 (0.70 to 0.80)	<0.001***

CCI, Charlson Comorbidity Index; ICG, integrative care group; LTCH, long-term care hospital; TKM, traditional Korean medicine; UCG, usual care group.

Risk of major medical complications

For the secondary outcomes, the ICG exhibited a significantly lower incidence of all evaluated medical complications. The reduced risk of these complications was evident in the unadjusted models and persisted significantly after all intermediate adjustments. For pneumonia, the most frequent complication, the fully adjusted model revealed a 28% reduced risk in the ICG (2920 vs 4439 events; aHR 0.72 (95% CI 0.69 to 0.76), p<0.001). Furthermore, integrative care was independently associated with decreased risks of other complications, including DVT (158 vs 255 events; aHR 0.71 (95% CI 0.58 to 0.88), p=0.001), pressure ulcers (2540 vs 3616 events; aHR 0.76 (95% CI 0.72 to 0.80), p<0.001) and UTIs (1587 vs 2312 events; aHR 0.75 (95% CI 0.70 to 0.80), p<0.001).

The detailed HR estimates for all covariates included in the fully adjusted models are provided in online supplemental table 2.

Sensitivity analyses

For the prespecified subgroup analyses, we applied the identical sequential modelling approach to each stratum (online supplemental table 3). Regarding the primary outcome of all-cause mortality, while the inverse association between integrative care and mortality remained statistically significant across most demographic and clinical subgroups, its magnitude varied. Compared with the overall matched cohort, the risk reduction was relatively attenuated in males (aHR 0.89 (95% CI 0.85 to 0.95)) and the oldest adults aged ≥85 years (aHR 0.85 (95% CI 0.82 to 0.89)), though still significant. When stratified by baseline comorbidity burden, the association with lower mortality was highly significant in patients with a CCI score of 0 (aHR 0.77 (95% CI 0.75 to 0.79)) but lost statistical significance in those with a higher comorbidity burden (CCI score ≥1).

A similar pattern of varying effect sizes was observed for the secondary outcomes of major medical complications. Although the significant risk reductions for pneumonia, pressure ulcers and UTIs were broadly maintained across most subgroups, the strength of the association differed. For instance, the inverse association with pneumonia was stronger in females (aHR 0.67 (95% CI 0.63 to 0.71)) and the 65–74 age group (aHR 0.59 (95% CI 0.49 to 0.70)), compared with males (aHR 0.82 (95% CI 0.75 to 0.90)) and the oldest old (≥85 years; aHR 0.83 (95% CI 0.78 to 0.90)). Furthermore, the inverse association with DVT exhibited substantial variation, losing statistical significance in several subgroups. In the fully adjusted models, the association was not significant in males, the 65–74 age group and Medical Aid beneficiaries. This attenuation likely reflects both the varying clinical profiles of these subpopulations and the limited statistical power resulting from lower event rates in smaller strata.

Within each fracture subtype, separate matching achieved adequate covariate balance (online supplemental table 4), and the ICG showed lower absolute incidence rates than the UCG for all outcomes. For all-cause mortality, the incidence rate per 100 person-years was 18.23 vs 22.76 for femur fractures and 15.77 vs 21.43 for vertebral fractures (online supplemental table 5). Consistent with these findings, the association with lower mortality was more pronounced in patients with wrist (aHR 0.70 (95% CI 0.61 to 0.80)) and shoulder fractures (aHR 0.73 (95% CI 0.64 to 0.84)) than in those with femur fractures (aHR 0.81 (95% CI 0.78 to 0.84)). For the secondary outcomes, the inverse association with pneumonia was likewise stronger in wrist fractures (aHR 0.58 (95% CI 0.45 to 0.75)) than in femur fractures (aHR 0.77 (95% CI 0.72 to 0.82)), whereas the association with DVT was not significant across any individual fracture subtype (online supplemental table 6). The forest plot for the fully adjusted models across all subgroup analyses is shown in figure 2.

Figure 2. Subgroup analyses of the fully adjusted HRs for all-cause mortality and major immobility-related complications. The usual care group served as the reference category. HRs and 95% CIs for each subgroup were estimated using the fully adjusted Cox proportional hazards regression model. CCI, Charlson Comorbidity Index; NE, not estimable.

In the treatment-frequency analysis, for all-cause mortality, the aHRs relative to the UCG were 0.83 (95% CI 0.80 to 0.85) for 1/28≤PDC<1/21 and 0.70 (95% CI 0.67 to 0.73) for 1/21≤PDC<1/14, whereas the estimate for PDC ≥1/14 was no longer statistically significant (0.95 (95% CI 0.87 to 1.05), p for trend <0.001). In the highest-frequency category, the associations were likewise non-significant for DVT, pressure ulcers and UTIs, whereas the reduced risk of pneumonia remained significant across all categories. Detailed estimates for all outcomes are provided in online supplemental table 7. The E-values for the fully adjusted point estimates, with the values for the confidence limit closest to the null in parentheses, were 2.12 (1.96) for pneumonia, 2.17 (1.53) for DVT, 1.96 (1.81) for pressure ulcers and 2.00 (1.81) for UTIs. For all-cause mortality, the E-values were 1.85 (1.77) when the HR was treated as a risk ratio and 1.63 (1.58) after accounting for the outcome frequency.

Discussion

In this population-based, propensity score-matched, retrospective cohort study using national claims data from 2014 to 2023, we investigated whether sustained integrative care, combining TKM with conventional medical care, is associated with lower mortality and reduced risk of major medical complications among older adults undergoing postfracture rehabilitation in Korean LTCHs. Our principal findings demonstrate that integrative care was significantly associated with a 21% lower risk of all-cause mortality and reduced incidences of major immobility-related complications, including pneumonia, DVT, pressure ulcers and UTIs. The observed associations varied across several strata. Specifically, lower risk estimates were observed among females, younger geriatric patients (65–74 years) and those with upper-extremity fractures, whereas the associations were attenuated in those with femur fractures and patients with a higher pre-existing comorbidity burden.

Our findings are broadly consistent with previous population-based studies investigating the use of traditional medicine in fracture-related cohorts. For instance, integrating acupuncture or herbal medicine into postoperative care has been associated with significantly lower mortality and readmission rates among patients with hip fractures.^{11 13} Similar associations with lower mortality and reduced hospitalisation risks have also been observed when traditional medicine was applied to patients with hyperthyroidism and osteoporosis.¹² The present study extends this body of literature by evaluating a nationwide cohort of older adults with diverse fracture types admitted to Korean LTCHs and by providing evidence on immobility-related complications.

Several hypotheses may potentially explain the observed associations, although these mechanisms were not directly evaluated in the present study. Clinical studies have demonstrated that acupuncture and electroacupuncture may alleviate postfracture pain, improve sleep quality and restore bowel function in geriatric rehabilitation settings.^34–36 Furthermore, recent evidence suggests that electroacupuncture modulates inflammatory cytokines, promotes angiogenesis and accelerates bone healing,³⁷ while cupping therapy improves local blood microcirculation and relieves musculoskeletal pain.³⁸ These effects could potentially reduce the physiological stress associated with trauma and facilitate earlier mobilisation and participation in rehabilitation, which may in turn be associated with lower risks of immobility-related complications such as pneumonia, DVT and pressure ulcers.^{10 24 26 27 37} However, these proposed mechanisms remain speculative in the context of the present claims-based observational study.

In the subgroup analyses, the observed associations were attenuated among patients with a higher baseline comorbidity burden. In these highly vulnerable individuals with severe underlying diseases, pre-existing conditions may contribute substantially to the risks of mortality and major medical complications.^{1 2} However, because formal interaction tests were not performed and baseline comorbidity may have been incompletely ascertained, these subgroup findings should be interpreted cautiously and should not be considered evidence that the association between integrative care and outcomes differs according to comorbidity burden.

This study has several strengths. First, the use of a nationwide claims database, encompassing all relevant patients over 10 years and including mortality data linked to the national death registry, ensures the high representativeness of our findings. Additionally, we used PSM and sequentially adjusted Cox proportional hazards models to improve comparability between the study groups with respect to measured covariates. Furthermore, by evaluating real-world integrative practice, we provide insights into postfracture recovery trajectories in LTCHs, particularly regarding immobility-related complications beyond all-cause mortality. Finally, we conducted extensive sensitivity analyses to examine the consistency of the observed associations across subgroups and alternative exposure definitions.

However, several limitations must be acknowledged. Formal interaction tests were not performed, so the subgroup findings should be regarded as exploratory rather than as evidence of effect modification. Also, adjustment for matched-pair and facility-level clustering, risk differences, and numbers needed to treat were not undertaken. Given the retrospective, observational nature of this study, residual confounding due to unmeasured clinical variables cannot be completely ruled out. For example, fracture severity and timing, surgical treatment, prefracture mobility, cognitive impairment, frailty, pain severity, rehabilitation intensity, nutritional status, baseline functional reserve and patient or family treatment preferences could not be captured, yet may influence both treatment allocation and prognosis. Even though the E-values suggested that weak confounding alone is unlikely to explain the findings, the observed associations may still partly reflect selection of patients who were healthier, more mobile or treated in better-resourced facilities, rather than the effects of integrative care alone.

In addition, because the classification of integrative care required patients to survive and remain hospitalised long enough to receive sustained treatment, immortal time bias cannot be completely ruled out. The exclusion of patients with hospital stays shorter than 30 days and those with sporadic TKM use may also have introduced selection bias by retaining a healthier and more stable subgroup. Although a 30-day landmark restriction was applied, this approach may not have fully eliminated these potential biases. Moreover, the ≥1 session per 28 days exposure threshold and the 30-day landmark were pragmatic choices, and alternative definitions could yield different exposure classifications, although the treatment-frequency analysis showed broadly consistent associations. Furthermore, the study population included various fracture subtypes with potentially different prognoses, rehabilitation pathways, treatment intensity and mortality risks. Although fracture-specific subgroup analyses were conducted, residual confounding related to differences in fracture phenotype may remain.

Because Fine-Gray subdistribution models were not fitted, the absolute cumulative-incidence risks were not characterised, and the findings for secondary outcomes should be interpreted accordingly. Information on race and ethnicity was not provided in the NHIS claims database. Given that the study population was predominantly Korean, this omission is unlikely to have introduced meaningful confounding. Additionally, because our database captures only LTCH claims and patients who develop acute conditions during their stay are typically transferred to acute-care hospitals, many diagnoses are recorded outside the LTCH claims. This likely contributed to the low mean CCI scores and to underascertainment of baseline comorbidity burden, leaving residual confounding despite PSM. The CCI-stratified subgroup findings should therefore be interpreted with particular caution. In addition, no complication-specific preindex washout periods were applied, and a single qualifying claims diagnosis was sufficient to define an outcome. Although diagnoses recorded during the first 30 days after the index admission were not counted as outcome events, some subsequently identified complications may have represented pre-existing or recurrent conditions rather than newly developed complications during rehabilitation. Furthermore, the claims-based outcome definitions were not independently validated in this study. Therefore, outcome misclassification cannot be excluded.

The exclusion of records with missing or invalid covariate values used for propensity-score matching may also have introduced selection bias. However, because only a small number of patients (n=531) were excluded, any influence on the findings is likely limited. Furthermore, the lack of granular treatment details in the administrative data limited our ability to evaluate the specific effects of individual interventions within the exposure group. Non-reimbursed medical services, such as traditional herbal medicine, were also inevitably excluded from the analysis. Finally, because this study is situated within the unique structure of the Korean long-term care system, caution is required when generalising these findings to countries with different healthcare environments.

Conclusions

In conclusion, this nationwide cohort study shows that sustained integrative care was associated with lower all-cause mortality and fewer major immobility-related complications among older adults undergoing postfracture rehabilitation in LTCHs. Even though the subgroup analyses suggested that the association differed according to patients’ baseline conditions, this apparent difference requires confirmation with formal interaction testing. Furthermore, to overcome the inherent limitations of administrative claims data, future large-scale prospective cohort studies that incorporate detailed clinical outcomes and precise baseline functional status are essential to establish optimised strategies for the clinical application of integrative medicine in geriatric fracture care.

Data availability statement

Data may be obtained from a third party and are not publicly available.