Characteristics and outcomes of non-critically ill extremely elderly patients with do-not-attempt-resuscitation orders admitted to a university hospital in Japan: a retrospective cohort study Authors: Sugihiro Hamaguchi, Takayuki Idaka, Ryutaro Usuki, Toru Naganuma, Sei Takahashi, Hiroaki Nakagawa, Tetsuro Aita, Satoshi Kanke Published: 2026 ABSTRACT BackgroundDo-not-attempt-resuscitation (DNAR) orders are commonly issued for patients of advanced age in the setting of severe illness with an anticipated risk of mortality. However, DNAR orders are also observed among extremely elderly patients who are not critically ill at admission. The characteristics and outcomes of such patients remain poorly understood. This study aimed to characterize these patients and to examine the association between DNAR status and discharge outcomes.MethodsWe retrospectively analyzed non-critically ill (National Early Warning Score [NEWS] of 4 or less) hospitalized patients aged 85 years or older admitted to a university hospital in Japan between 2022 and 2024 using electronic medical record data. Patients with and without DNAR orders were compared in terms of age, sex, activities of daily living (ADL), disease severity (NEWS), primary diagnoses, comorbidities, laboratory data, and discharge outcomes using univariable analyses. A Fine‒Gray competing-risk model examined the association between DNAR orders and transfer to post-acute care facilities, with discharge to home and death as competing events.ResultsAmong the 1,230 eligible patients, 72 (5.9%) had DNAR orders. Patients with DNAR orders were older, had lower ADL, and had higher NEWS at admission. Primary diagnoses and comorbidities showed no consistent pattern between the two groups. Laboratory findings at admission showed lower hemoglobin and albumin levels and higher C-reactive protein levels in the DNAR group. Patients with DNAR orders were less likely to be discharged home, more likely to be transferred to post-acute care facilities or die in hospital, and had longer hospital stays. DNAR orders were associated with a higher probability of transfer to post-acute facilities (subdistribution hazard ratio 2.11; 95% confidence interval 1.21–3.69).ConclusionsAmong non-critically ill extremely elderly hospitalized patients, those with DNAR orders were older, had lower ADL, and had higher disease severity. The presence of DNAR orders was associated with unfavorable discharge outcomes independent of age, ADL, and disease severity.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12877-026-07758-3. Background A do-not-attempt-resuscitation (DNAR) order is a formally documented medical decision that precludes the initiation of cardiopulmonary resuscitation (CPR) in the event of cardiopulmonary arrest, including chest compressions, defibrillation, and advanced airway management [1]. The specific scope of DNAR orders may vary across countries and institutions. In some settings, limitations on interventions such as intubation may be addressed separately (e.g., do-not-intubate orders). In principle, DNAR orders are intended to respect the patient’s preferences and to avoid resuscitation attempts that are unlikely to restore meaningful survival or may be medically non-beneficial [2]. Previous studies have reported higher mortality among severely ill patients with DNAR orders [3–6]. Importantly, a DNAR order applies only to resuscitation attempts and does not limit appropriate clinical management, including diagnostic or therapeutic procedures and pharmacologic treatments [7]. Thus, DNAR orders should not be interpreted as instructions to withhold necessary treatment. However, several studies have shown that DNAR orders are frequently issued without sufficient discussion among patients, their families, and the healthcare team, and that clinicians’ judgment or institutional culture may substantially influence the decision-making process rather than the patient’s own preferences [8–10]. Concerns have also been raised that the presence of a DNAR order may inadvertently lead to the withholding or delay of clinically indicated treatments that the patient would otherwise receive, resulting in unintended reductions in the quality or intensity of necessary care [11]. Although DNAR orders are generally applied to critically ill patients, they may also be assigned to extremely elderly patients with relatively mild conditions [12]. Such decisions may sometimes reflect patients’ preferences regarding life-sustaining treatment. However, in Japan, particularly in acute-care tertiary hospitals where the primary focus is life-saving treatment, DNAR orders are generally not routinely discussed for patients who are not expected to experience cardiopulmonary arrest during hospitalization, as advance care planning is often initiated after clinical deterioration [13], although practices may vary across physicians and institutions. In such cases, the decision may be influenced by clinicians’ subjective impressions or institutional routines rather than a clear clinical rationale. In some settings, discussions about DNAR orders may be initiated by healthcare providers as a precaution, and such discussions may not always be based on clearly established patient preferences or predefined treatment goals [14, 15]. Therefore, when DNAR orders are issued for extremely elderly patients with relatively mild illness, there is concern that the order itself may negatively influence the care for non-critically ill patients who are not expected to deteriorate during hospitalization. However, little is known about the characteristics of extremely elderly hospitalized patients with non-critical illness who receive DNAR orders. Among hospitalized older adults, the discharge destination is a particularly important outcome because a substantial proportion of elderly patients are discharged to other care settings such as post-acute care facilities rather than directly home after acute hospitalization [16]. However, it remains unclear whether and how DNAR orders influence discharge outcomes among extremely elderly patients with non-critical illness. This study aimed to describe the characteristics of these patients admitted to a university hospital in Japan and to investigate the associations between DNAR orders and patients’ discharge outcomes. Methods Study design and setting This retrospective cohort study used electronic medical record data from Fukushima Medical University Hospital. We identified all patients hospitalized between January 2022 and December 2024 and constructed a study database by extracting relevant clinical information. Fukushima Medical University Hospital, located in Fukushima Prefecture, Japan, is the only university hospital in the prefecture and functions as a tertiary referral center with 778 beds, admitting approximately 16,000 patients annually. The hospital receives a broad mix of patients, including those referred from community clinics and hospitals for specialized evaluation, as well as individuals arriving via emergency medical services or transferred from other facilities. Patients who are unable to return home after acute care are typically discharged to post-acute or long-term care hospitals within the regional network. As in many tertiary referral university hospitals in Japan, some patients are also referred from regional acute-care or subacute-care hospitals for specialized evaluation or treatment within the regional healthcare network. Although many such patients are ultimately discharged home, patients who cannot achieve sufficient recovery for direct discharge home are commonly discharged to post-acute care hospitals or back to referring institutions for continued care and rehabilitation. Clinical data were automatically extracted from the institutional electronic medical record system by the hospital’s medical information department according to predefined extraction criteria. The study investigators were not involved in the data extraction process. All variables, including demographic characteristics, clinical parameters, and outcomes, were identified from structured electronic medical record data fields based on prespecified definitions. After extraction, the dataset was de-identified using CoNaxs (4DIN Ltd., Tokyo, Japan), an anonymization software tool, and subsequently analyzed within a secure cloud-based data analysis platform (SIMPRESEARCHⓇ, 4DIN Ltd., Tokyo, Japan). No manual chart review was conducted; therefore, reviewer training, blinding, and interrater agreement assessment were not applicable. Study population We included patients aged 85 years or older who were hospitalized during the study period and met criteria for non-critical illness. The age 85 years is a commonly defined cut-off for oldest-old individuals in geriatric and epidemiologic research [17]. Clinical severity at presentation was assessed using the National Early Warning Score (NEWS), a standardized composite score incorporating respiratory rate, oxygen saturation, use of supplemental oxygen, temperature, blood pressure, pulse rate, and level of consciousness. The total NEWS ranges from 0 to 20 points and is used to estimate the short-term risk of clinical deterioration; scores of 0–4 indicate low risk according to the standardized scale [18]. We defined non-critically ill patients as those with a NEWS of 0–4 at admission, corresponding to the low clinical risk category in the original NEWS risk stratification. Scores 5 or higher are categorized as medium to high risk and warrant increased clinical monitoring. Therefore, we restricted the cohort to patients with clearly low initial severity. In addition, according to the NEWS guideline, patients with a single parameter scoring 3 should not be classified as low risk, based on expert opinion. However, the Japanese validation study assessed NEWS using only the aggregate score and did not separately evaluate the prognostic impact of a single parameter score 3 [19]. In addition, an observational study has reported that a single parameter score of 3 alone does not confer a risk equivalent to that associated with an aggregate NEWS of 5 [20]. When NEWS was measured multiple times on the day of admission, the earliest recorded measurement was used to calculate the score. To ensure that only patients with mild clinical conditions were analyzed, we excluded those admitted to the intensive care unit (ICU) or high care unit (HCU) at the time of hospital admission. Variables The explanatory variable of interest was the presence or absence of a DNAR order, which was formally documented in the electronic medical records when it was issued. The DNAR document in the hospital was a standardized written form that explained that CPR would not be performed in the event of cardiac or respiratory arrest when recovery was considered highly unlikely. DNAR orders were issued during hospitalization based on informed consent obtained from the patient. If the patient was unable to sign, a family member provided the signature on the patient’s behalf. The form was signed by the attending physician who provided the explanation and by a nurse who was present during the informed consent discussion. All other variables were treated as covariates. Demographic information consisted of age and sex. Disease severity at admission was assessed using the NEWS, and functional status was evaluated with the Katz Index of Independence in Activities of Daily Living (ADL) [21], which measures six basic functions: bathing, dressing, toileting, transferring, continence, and feeding. It yields a score ranging from 0, indicating complete dependence, to 6, indicating full independence. Primary diagnoses at admission were selected and categorized using major chapters of the International Classification of Diseases, Tenth Revision (ICD-10) to provide a standardized and comprehensive classification across heterogeneous clinical conditions. Those included neoplasms, cardiovascular diseases, eye disorders, gastrointestinal disorders, trauma or orthopedic conditions, skin disorders, and COVID-19. The comorbidities included diabetes, chronic heart failure, malignancies, end-stage kidney disease, and dementia. Comorbid conditions were identified using ICD-10 codes recorded at admission for the index hospitalization. The laboratory data obtained on the day of admission included hemoglobin concentration, platelet count, serum albumin level, and C-reactive protein (CRP). When multiple measurements were available on the day of admission, the first recorded value was used for analysis. Because previous studies specifically examining predictors of discharge to post-acute care facilities were limited, laboratory variables were selected based on prior evidence linking them to mortality in older adults [22–25], whereas comorbidities were selected a priori based on their clinical relevance to adverse outcomes in older patients. In addition, to account for overall comorbidity burden and to assess the robustness of the results, the Charlson Comorbidity Index (CCI) was calculated using the ICD-10 coding algorithm developed by Quan et al. [26] and used in a sensitivity analysis. The ICD-10 codes used to define the primary diagnoses and comorbidities are provided in Supplementary Table 1; Additional file 1. Outcomes The outcomes were the three discharge dispositions from the index hospitalization: discharge to home, transfer to post-acute care facilities, and in-hospital death. Among these, transfer to post-acute care facilities was the main outcome of interest because it reflects situations in which patients cannot be directly discharged home and require ongoing medical care. In our setting, this outcome often reflects inability to achieve direct discharge home after tertiary hospital treatment, including among patients referred from other acute-care or subacute-care institutions. Because the study population consisted of patients with mild clinical severity and in-hospital death was expected to be relatively uncommon, transfer to post-acute care facilities was considered the most clinically relevant outcome. A post-acute care facility was defined as any healthcare institution with full-time physicians to which patients were transferred when direct discharge to home was not feasible. This included chronic care hospitals, long-term care hospitals, skilled nursing facilities, rehabilitation hospitals, and hospice facilities. In addition, the length of hospital stay, defined as the number of days from admission to discharge, was also evaluated. Ethics This study was approved by the Ethics Committee of Fukushima Medical University (approval number: REC2024-148). Because the study used anonymized data extracted from electronic medical records, the requirement for individual informed consent was waived. In accordance with institutional policy and national ethical guidelines, information about the study was publicly disclosed to provide patients with the opportunity to opt out. All procedures were conducted in compliance with the Declaration of Helsinki and relevant local regulations on the protection of personal information. Bias, sample size, and missing data handling Because this was a retrospective cohort study using electronic medical record data, no a priori sample size calculation was performed; all eligible patients hospitalized during the study period were included. For the multivariable analysis (Fine‒Gray subdistribution hazard regression analysis), we limited the number of covariates in each model relative to the number of primary events, in accordance with simulation evidence suggesting that the number of events per variable should be considered when assessing model performance. Specifically, previous work has shown that accurate estimation of regression coefficients and associated quantities generally requires between 10 and 40 primary events per covariate in Fine‒Gray models, depending on the type of variables included in the model [27]. In the present study, we prespecified the inclusion of 15 covariates. To reduce selection bias, consecutive patients who met the inclusion criteria were enrolled. Information bias was minimized by relying on routinely collected electronic medical record data with standardized measurement procedures. Differences in baseline characteristics were addressed by adjusting for relevant covariates in the Fine‒Gray models. Given the lack of prior studies investigating transfer to post-acute care facilities as a discharge outcome among mildly ill hospitalized older adults, these covariates were selected based on predictors that have been reported to be associated with mortality outcomes in previous studies, including age, ADL, and disease severity, as well as additional factors considered clinically relevant. If the proportion of missing data was less than 10% [28] and no strong associations between missingness and observed covariates were identified, a complete-case analysis was conducted. Statistical analysis Continuous variables were summarized as medians with interquartile ranges (IQRs), and categorical variables as counts and percentages. Differences between groups were assessed using the Wilcoxon rank-sum test for continuous variables and the chi-square test or Fisher’s exact test for categorical variables, as appropriate. The three discharge outcomes were discharge to home, transfer to post-acute care facilities, and in-hospital death. The cumulative incidence functions for these outcomes were estimated according to DNAR status while accounting for competing risks. Associations between DNAR status and each outcome were evaluated using Fine‒Gray subdistribution hazard models, which assumed proportional subdistribution hazards, with transfer to post-acute care facilities as the primary outcome of interest. For transfer to post-acute care facilities, the model included DNAR status and the following covariates: age, sex, NEWS, Katz index, primary diagnosis categories (neoplasms, cardiovascular disorders, eye disorders, gastrointestinal disorders, and trauma or orthopedic disorders, with COVID-19 as a separate category) and comorbid conditions (chronic heart failure, malignancies, dementia, and end-stage kidney disease). Laboratory variables were summarized descriptively but were not included in the multivariable models because a substantial proportion of patients did not undergo these tests. The availability of each laboratory test was therefore reported for each group. Because the study period overlapped with the COVID-19 pandemic, COVID-19 was treated as a separate diagnostic category and included in the multivariable models to account for its potential influence on DNAR decisions and discharge outcomes. For discharge to home and in-hospital death, Fine‒Gray models were also fitted as supplementary analyses. In addition, three sensitivity analyses were conducted. First, because a single NEWS parameter scoring 3 may indicate potential clinical instability despite a total NEWS of 0–4, an additional analysis was performed after excluding such patients. This analysis was restricted to patients without missing covariates. Second, the Fine–Gray model was refitted including the Charlson Comorbidity Index (CCI) instead of individual comorbid conditions to account for overall comorbidity burden. Third, to address the potential issue of multiple comparisons across the three discharge outcomes, a sensitivity analysis using more conservative confidence intervals corresponding to a Bonferroni correction (98.33% confidence intervals) was performed. All statistical tests were two-tailed, and a p-value of less than 0.05 was considered statistically significant. All analyses were performed using R version 4.1.3 (R Foundation for Statistical Computing, Vienna, Austria). This study was reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines [29]. Results Study population and flow A total of 1,422 patients aged 85 years or older with a NEWS of 4 or less were hospitalized during the study period. After excluding 192 patients admitted to the ICU or HCU, 1,230 patients were eligible for the final analysis (Fig. 1). Among the analyzed cohort, 72 patients (5.9%) had a documented DNAR order during hospitalization, whereas 1,158 patients (94.1%) did not.Fig. 1Study flow of patient selection. Among 1,230 non-critical extremely elderly patients eligible for the analysis, 72 (5.9%) had DNAR orders. NEWS, National Early Warning Score; ICU, intensive care unit; HCU, high care unit; DNAR, do-not-attempt-resuscitation Study flow of patient selection. Among 1,230 non-critical extremely elderly patients eligible for the analysis, 72 (5.9%) had DNAR orders. NEWS, National Early Warning Score; ICU, intensive care unit; HCU, high care unit; DNAR, do-not-attempt-resuscitation Baseline characteristics according to DNAR status The baseline characteristics of patients with and without DNAR orders are summarized in Table 1.Table 1Characteristics of patients with or without DNAR ordersOverallWith DNARWithout DNARSMDN = 1230N = 72N = 1158Age, median [IQR]87.0[86.0, 90.0]89.5[87.0, 93.0]87.0[86.0, 89.0]0.59Female641(52.1)38(52.8)603(52.1)0.01Katz Index, median [IQR]*6[2, 6]0[0, 2]6[3, 6]1.60Days from admission to DNAR order, median [IQR]1[1, 4.3]1[1, 4.3]NANANAPrimary diagnosis for admission, n (%) Neoplasms333(27.1)18(25.0)315(27.2)0.05 Cardiovascular diseases249(20.2)6(8.3)243(21.0)0.36 Eye disorders244(19.8)0(0.0)244(21.1)0.73 Trauma/Orthopedic disorders122(10.0)4(5.6)118(10.2)0.17 GI disorders83(6.7)15(20.8)68(5.9)0.45 Skin disorders30(2.4)6(8.3)24(2.1)0.28 COVID-1922(1.8)11(15.3)11(0.9)0.54Comorbid diseases, n (%) Diabetes mellitus167(13.6)10(13.9)157(13.6)0.01 Chronic heart failure164(13.3)5(6.9)159(13.7)0.22 Malignancies39(3.2)9(12.5)30(2.6)0.38 End-stage kidney diseases33(2.7)4(5.6)29(2.5)0.16 Dementia13(1.1)2(2.8)11(0.9)0.14CCI score, n (%)0.45 0815(66.3)44(61.1)771(66.6) 1273(22.2)11(15.3)262(22.6) 297(7.9)7(9.7)90(7.8) 326(2.1)5(6.9)21(1.8) 43(0.2)0(0.0)3(0.3) 54(0.3)1(1.4)3(0.3) 612(1.0)4(5.6)8(0.7)NEWS, median [IQR]0.5[0.0, 2.0]1[0.0, 3.0]0[0.0, 1.0]0.49Laboratory tests, median [IQR]* Hemoglobin (g/dL)11.6[10.2, 12.6]10.5[9.1, 11.5]11.7[10.3, 12.8]0.64 unmeasured, n (%)568(46.2)3(4.2)565(48.8) Platelet (/mm3)200[158, 257]199[154, 278]200[158, 253]0.10 unmeasured, n (%)568(46.2)3(4.2)565(48.8) Albumin (g/dL)3.6[3.1, 3.9]2.8[2.4, 3.3]3.7[3.2, 4.0]1.20 unmeasured, n (%)611(49.7)8(11.1)603(52.1) CRP (mg/dL)0.25[0.03, 1.75]5.4[1.00, 9.81]0.17[0.02, 1.17]0.91 unmeasured, n (%)579(47.1)3(4.2)576(49.7)DNAR Do-not-attempt-resuscitation, SMD Standardized mean difference, IQR Interquartile range, NEWS National Early Warning Score, GI Gastrointestinal, CRP C-reactive protein*Number of cases: Katz Index, N = 1,109; Albumin, N = 619; Hemoglobin, N = 662; Platelet, N = 662 CRP, N = 651 Table 1Characteristics of patients with or without DNAR ordersOverallWith DNARWithout DNARSMDN = 1230N = 72N = 1158Age, median [IQR]87.0[86.0, 90.0]89.5[87.0, 93.0]87.0[86.0, 89.0]0.59Female641(52.1)38(52.8)603(52.1)0.01Katz Index, median [IQR]*6[2, 6]0[0, 2]6[3, 6]1.60Days from admission to DNAR order, median [IQR]1[1, 4.3]1[1, 4.3]NANANAPrimary diagnosis for admission, n (%) Neoplasms333(27.1)18(25.0)315(27.2)0.05 Cardiovascular diseases249(20.2)6(8.3)243(21.0)0.36 Eye disorders244(19.8)0(0.0)244(21.1)0.73 Trauma/Orthopedic disorders122(10.0)4(5.6)118(10.2)0.17 GI disorders83(6.7)15(20.8)68(5.9)0.45 Skin disorders30(2.4)6(8.3)24(2.1)0.28 COVID-1922(1.8)11(15.3)11(0.9)0.54Comorbid diseases, n (%) Diabetes mellitus167(13.6)10(13.9)157(13.6)0.01 Chronic heart failure164(13.3)5(6.9)159(13.7)0.22 Malignancies39(3.2)9(12.5)30(2.6)0.38 End-stage kidney diseases33(2.7)4(5.6)29(2.5)0.16 Dementia13(1.1)2(2.8)11(0.9)0.14CCI score, n (%)0.45 0815(66.3)44(61.1)771(66.6) 1273(22.2)11(15.3)262(22.6) 297(7.9)7(9.7)90(7.8) 326(2.1)5(6.9)21(1.8) 43(0.2)0(0.0)3(0.3) 54(0.3)1(1.4)3(0.3) 612(1.0)4(5.6)8(0.7)NEWS, median [IQR]0.5[0.0, 2.0]1[0.0, 3.0]0[0.0, 1.0]0.49Laboratory tests, median [IQR]* Hemoglobin (g/dL)11.6[10.2, 12.6]10.5[9.1, 11.5]11.7[10.3, 12.8]0.64 unmeasured, n (%)568(46.2)3(4.2)565(48.8) Platelet (/mm3)200[158, 257]199[154, 278]200[158, 253]0.10 unmeasured, n (%)568(46.2)3(4.2)565(48.8) Albumin (g/dL)3.6[3.1, 3.9]2.8[2.4, 3.3]3.7[3.2, 4.0]1.20 unmeasured, n (%)611(49.7)8(11.1)603(52.1) CRP (mg/dL)0.25[0.03, 1.75]5.4[1.00, 9.81]0.17[0.02, 1.17]0.91 unmeasured, n (%)579(47.1)3(4.2)576(49.7)DNAR Do-not-attempt-resuscitation, SMD Standardized mean difference, IQR Interquartile range, NEWS National Early Warning Score, GI Gastrointestinal, CRP C-reactive protein*Number of cases: Katz Index, N = 1,109; Albumin, N = 619; Hemoglobin, N = 662; Platelet, N = 662 CRP, N = 651 Characteristics of patients with or without DNAR orders DNAR Do-not-attempt-resuscitation, SMD Standardized mean difference, IQR Interquartile range, NEWS National Early Warning Score, GI Gastrointestinal, CRP C-reactive protein *Number of cases: Katz Index, N = 1,109; Albumin, N = 619; Hemoglobin, N = 662; Platelet, N = 662 CRP, N = 651 Patients with DNAR orders tended to be older and had lower ADL, as assessed by the Katz Index, compared with those without DNAR orders. With respect to primary diagnoses at admission, there were some differences between the two groups but these differences did not show a consistent or interpretable pattern across diagnostic categories. Notably, eye disorders were observed only in patients without DNAR orders. These admissions primarily represented short-term hospitalizations for ophthalmologic procedures such as cataract or glaucoma surgery. COVID-19, however, was substantially more prevalent among patients with DNAR orders. Most comorbid conditions were similarly distributed between the two groups, although malignancies were more prevalent among patients with DNAR orders, despite no difference in the prevalence of malignant neoplasms as the primary diagnosis at admission. Although all patients included in the study had a NEWS of 4 or less at admission, patients with DNAR orders had higher NEWS values compared with those without DNAR orders. At admission, patients with DNAR orders were more likely to be anemic, hypoalbuminemic, and to have higher levels of systemic inflammation than those without DNAR orders. Laboratory data were not available for all patients, and analyses were therefore based on available data. Information on the timing of DNAR orders was also examined. DNAR orders were typically established early after admission (median, 1 day). Discharge outcomes and length of hospital stay by DNAR status Discharge outcomes and length of hospital stay according to DNAR status are summarized in Table 2. Most patients were discharged home, while in-hospital death was rare (1.9%). Patients with DNAR orders were more likely to be transferred to post-acute care facilities and to die in hospital, and were less likely to be discharged home. They also had a longer length of hospital stay.Table 2Discharge outcomes by DNAR statusOverallWith_DNARWithout_DNARp-valueN = 1230N = 72N = 1158Discharge home, n (%)103083.72534.7100586.8 < 0.001Transfer to post-acute care facilities, n (%)17714.43447.214312.3 < 0.001Death, n (%)231.91318.1100.9 < 0.001Length of stay, median [IQR]7.0[4.0, 15.0]17.0[7.8, 35.3]7.0[3.0, 14.0] < 0.001DNAR Do-not-attempt-resuscitation, IQR Interquartile range Table 2Discharge outcomes by DNAR statusOverallWith_DNARWithout_DNARp-valueN = 1230N = 72N = 1158Discharge home, n (%)103083.72534.7100586.8 < 0.001Transfer to post-acute care facilities, n (%)17714.43447.214312.3 < 0.001Death, n (%)231.91318.1100.9 < 0.001Length of stay, median [IQR]7.0[4.0, 15.0]17.0[7.8, 35.3]7.0[3.0, 14.0] < 0.001DNAR Do-not-attempt-resuscitation, IQR Interquartile range Discharge outcomes by DNAR status DNAR Do-not-attempt-resuscitation, IQR Interquartile range Cumulative incidence of outcomes The cumulative incidence functions for transfer to post-acute care facilities, discharge to home, and in-hospital death according to DNAR status are shown in Fig. 2. Patients with DNAR orders had a consistently higher cumulative incidence of transfer to post-acute care facilities throughout hospitalization. In contrast, the cumulative incidence of discharge to home increased more rapidly among patients without DNAR orders and reached a plateau earlier than that observed among patients with DNAR orders. The cumulative incidence of in-hospital death was also higher among patients with DNAR orders over time.Fig. 2Cumulative incidence of outcomes. DNAR orders were associated with higher cumulative incidences of transfer to post-acute care facilities and in-hospital death, whereas patients without DNAR orders showed a faster rise and earlier plateau in home discharge. DNAR, do-not-attempt-resuscitation Cumulative incidence of outcomes. DNAR orders were associated with higher cumulative incidences of transfer to post-acute care facilities and in-hospital death, whereas patients without DNAR orders showed a faster rise and earlier plateau in home discharge. DNAR, do-not-attempt-resuscitation Competing risk analysis for transfer to post-acute care facilities We conducted a competing risk analysis using the Fine‒Gray subdistribution hazard model to examine factors associated with transfer to post-acute care facilities, with discharge to home and in-hospital death treated as competing events. Among the 1,230 patients included in the analysis, missing data were present in 121 patients (9.8%), and missingness occurred exclusively for the Katz Index. The Katz Index consists of six component items and was considered missing if any item was undocumented. This pattern suggested measurement-specific missingness related to documentation practices rather than patient characteristics, and missingness was consistent with a missing completely random mechanism. Therefore, a complete-case analysis was performed, resulting in a total of 1,109 patients. Laboratory variables were not included in the multivariable models because approximately half of the laboratory data were missing. The primary outcome (transfer to post-acute care facilities) occurred in 177 patients, corresponding to approximately 12 events per variable in the primary model with 15 covariates. As shown in Table 3, the presence of a DNAR order was associated with a significantly higher subdistribution hazard of transfer to post-acute care facilities (subdistribution hazard ratio [SHR], 2.11; 95% confidence interval [CI], 1.21‒3.69), even after adjustment for age, ADL, NEWS, and other clinical covariates.Table 3Fine‒Gray analysis for “Transfer to post-acute care facilities”: complete case analysis of 1,109 casesVariablesUnivariable analysisMultivariable analysisSHR95%CISHR95%CIWith DNAR5.293.56—7.862.111.21—3.69Basic status Age1.051.00—1.110.950.89—1.02 Female0.560.40—0.790.700.48—1.03 NEWS (1 point increase)1.411.25—1.601.090.95—1.25 Katz index (1 point increase)*0.640.60—0.690.700.64—0.76Primary diagnosis for admission Neoplasms0.520.34—0.791.010.57—1.80 Cardiovascular disorders1.210.82—1.782.031.13—3.64 Eye disorders0.030.00—0.180.080.01—0.61 GI disorders1.640.89—3.031.320.62—2.81 Trauma/Orthopedic disorders4.693.27—6.723.732.11—6.60 COVID-194.362.02—9.401.730.76—3.90Comorbid diseases Chronic heart failure1.170.75—1.830.890.54—1.47 Malignancies1.940.97—3.882.070.99—4.35 Dementia2.260.74—6.901.110.33—3.72 End-stage kidney diseases2.651.39—5.051.180.67—2.10SHR Subdistribution hazard ratio, CI Confidence interval, DNAR Do-not-attempt-resuscitation, NEWS National Early Warning Score*Number of cases: Katz Index, N = 1,109 Table 3Fine‒Gray analysis for “Transfer to post-acute care facilities”: complete case analysis of 1,109 casesVariablesUnivariable analysisMultivariable analysisSHR95%CISHR95%CIWith DNAR5.293.56—7.862.111.21—3.69Basic status Age1.051.00—1.110.950.89—1.02 Female0.560.40—0.790.700.48—1.03 NEWS (1 point increase)1.411.25—1.601.090.95—1.25 Katz index (1 point increase)*0.640.60—0.690.700.64—0.76Primary diagnosis for admission Neoplasms0.520.34—0.791.010.57—1.80 Cardiovascular disorders1.210.82—1.782.031.13—3.64 Eye disorders0.030.00—0.180.080.01—0.61 GI disorders1.640.89—3.031.320.62—2.81 Trauma/Orthopedic disorders4.693.27—6.723.732.11—6.60 COVID-194.362.02—9.401.730.76—3.90Comorbid diseases Chronic heart failure1.170.75—1.830.890.54—1.47 Malignancies1.940.97—3.882.070.99—4.35 Dementia2.260.74—6.901.110.33—3.72 End-stage kidney diseases2.651.39—5.051.180.67—2.10SHR Subdistribution hazard ratio, CI Confidence interval, DNAR Do-not-attempt-resuscitation, NEWS National Early Warning Score*Number of cases: Katz Index, N = 1,109 Fine‒Gray analysis for “Transfer to post-acute care facilities”: complete case analysis of 1,109 cases SHR Subdistribution hazard ratio, CI Confidence interval, DNAR Do-not-attempt-resuscitation, NEWS National Early Warning Score *Number of cases: Katz Index, N = 1,109 As supplementary analyses, competing risk analyses using the Fine‒Gray model were also performed for in-hospital death and discharge to home, with the remaining two outcomes considered competing events, respectively. For the analysis of in-hospital death, only 23 deaths occurred during hospitalization. Therefore, to avoid model overfitting, the multivariable model included DNAR status and was adjusted only for age, sex, and the Katz Index. In this analysis, DNAR orders were associated with an increased subdistribution hazard of in-hospital death (SHR, 9.04; 95% CI, 2.17–25.79) (Supplementary Table 2; Additional file 3). In contrast, in the competing risk analysis for discharge to home, DNAR orders were associated with a lower subdistribution hazard of discharge to home (SHR, 0.26; 95% CI, 0.17‒0.41) after adjustment for clinical covariates (Supplementary Table 3; Additional file 3). Sensitivity analyses The results of the sensitivity analyses were consistent with those of the primary analysis. A total of 62 patients, 7 patients (4.7%) in the DNAR group and 55 patients (9.7%) in the non-DNAR group, had at least one parameter scoring 3. After excluding 62 patients with a single NEWS parameter scoring 3 and restricting the analysis to cases without missing covariates (n = 1,060), the association between DNAR status and transfer to post-acute care facilities remained similar to that in the primary analysis (Supplementary Table 4; Additional file 4). When the Charlson Comorbidity Index was included instead of individual comorbidities, the results were also unchanged (Supplementary Table 5; Additional file 4). In addition, analyses using more conservative confidence intervals corresponding to a Bonferroni correction (98.33% confidence intervals) yielded essentially identical conclusions (Supplementary Table 6; Additional file 4). Discussion In this retrospective cohort study of non-critically ill hospitalized patients aged 85 years or older, we found that although DNAR orders were rarely issued (5.9%), patients with DNAR orders were older, had lower ADL, and had greater disease severity at admission. Most DNAR orders were established early after admission. The primary diagnoses and comorbidities overall revealed no consistent pattern between groups. Eye disorders were observed only in patients without DNAR orders. These admissions mainly represented short-term hospitalizations for ophthalmologic procedures such as cataract or glaucoma surgery, for which DNAR orders are generally not obtained. In contrast, COVID-19 was more common in the DNAR group, which may reflect clinicians’ anticipation of potential deterioration among extremely elderly patients with COVID-19 during hospitalization. The laboratory findings at admission showed lower hemoglobin and albumin levels and higher CRP levels in the DNAR group. However, laboratory tests were not uniformly performed because they were ordered according to clinical judgment. Consequently, missing laboratory data likely reflected the absence of clinical indications rather than random missingness. Consistent with this interpretation, laboratory tests were less frequently performed in patients without DNAR orders. The laboratory values presented in Table 1 should therefore be interpreted descriptively. Patients with DNAR orders were more likely to be transferred to post-acute care facilities or die during hospitalization and were less likely to be discharged home. DNAR status was independently associated with transfer to post-acute care facilities. A DNAR order is a code status documented by clinicians for patients with little or no chance of recovery at the time of cardiac arrest to avoid non-beneficial CPR that the patient or family does not wish to receive [1]. Previous studies have identified factors associated with DNAR orders, including advanced age, poor ADL, and greater disease severity [8, 30, 31], consistent with our findings. DNAR orders have also been associated with outcomes such as mortality [32–34]. A major concern is that DNAR orders may influence clinical decisions beyond cardiopulmonary arrest and lead to the withholding of otherwise appropriate treatment [16, 35–37]. Although most previous studies have focused on critically ill patients, our findings in non-critically ill elderly patients also demonstrated that DNAR status was independently associated with transfer to post-acute care facilities, with similar trends observed for in-hospital death. The persistence of this association after adjustment for age, ADL, and disease severity suggests a potential influence of treatment limitations related to DNAR orders. Moreover, the fact that DNAR orders were established early after admission despite the patients having mild illness suggests that they may have been obtained for reasons unrelated to the patients’ clinical condition after admission. An exploratory study reported that DNAR orders were often documented early during hospitalization among very elderly patients [12], suggesting that such decisions may sometimes be established as a precautionary measure. From a health policy perspective, DNAR decisions should be incorporated into broader care planning frameworks such as advance care planning [37]. Several established frameworks, including the Physician Orders for Life Sustaining Treatment (POLST) [38], the Universal Form of Treatment Options (UFTO) [39], and the Recommended Summary Plan for Emergency Care and Treatment (ReSPECT) [40], provide practical models for integrating resuscitation decisions into comprehensive care plans and may help ensure that DNAR orders are applied appropriately without unnecessarily limiting care. A major strength of this study is its focus on mildly ill patients within an extremely elderly population, a group that is more likely to receive DNAR orders based on age but less likely based on clinical severity. This design allowed us to highlight the fundamental features and outcomes of “just in case” DNAR orders. On the other hand, this study has several limitations. First, it was conducted at a single center, which may limit generalizability. Future multicenter studies that include different types of institutions are needed to confirm our results. Second, approximately half of the laboratory variables at admission were missing, which prevented their inclusion in the multivariable analyses. In such situations, multiple imputation is often considered. However, given the high proportion of missing data and the possibility that missing laboratory data reflected the absence of clinical indications, the use of multiple imputation could have introduced substantial uncertainty and potentially distorted the results. Moreover, in routine clinical practice, clinicians do not typically rely on admission laboratory values alone when determining discharge disposition. For these reasons, laboratory variables were not included in the multivariable analyses. Third, in the supplementary analysis of in-hospital death using a competing risk model, the number of deaths was small possibly because the study population consisted of mildly ill patients. As a result, only age, sex, and the Katz Index were included as covariates, in addition to DNAR status. Because of this limitation in the number of variables, several potential covariates, including primary admission diagnoses, were not included in the model, and residual confounding may remain. Future studies with larger sample sizes, obtained from multicenter datasets or extended electronic medical record periods, will be needed to allow the inclusion of additional covariates. Finally, because this study used an anonymized database derived from structured hospital data, several clinically important factors could not be assessed. In particular, information on treatment limitations other than CPR, palliative care involvement, prior advance care planning (ACP), socioeconomic status, and detailed aspects of the DNAR decision-making process were not available. In addition, frailty could not be directly measured because validated frailty indices were not available in the database, although functional status at admission was assessed using the Katz Index. These unmeasured factors may have influenced both DNAR decisions and patient outcomes. Therefore, the observed associations between DNAR status and outcomes should be interpreted with caution, as they may partly reflect confounding by indication and residual confounding. To clarify the assumed causal structure and explicitly illustrate these potential sources of unmeasured confounding, we constructed a directed acyclic graph and incorporated these factors as unmeasured variables (Supplementary Fig. 1; Additional file 2). In Japan, the implementation of ACP remains limited. Detailed data on its prevalence in the general population are lacking, but fewer than 30% of healthcare professionals (27–28%) have reported ever being involved in ACP discussions [41]. The provision of early palliative care is also uncommon. According to one study, less than 5% of families reported early referral to specialist palliative care services [42]. Therefore, it is likely that many of the study participants would not have undergone ACP discussions or received palliative care consultation prior to admission. Conclusions In our study of DNAR orders among non-critically ill patients aged 85 years or older, a population more likely to receive DNAR orders because of advanced age but less likely because of mild illness, the presence of DNAR orders was associated with unfavorable discharge outcomes including transfer to post-acute care facilities independent of age, ADL, and disease severity. These findings may reflect the insufficient clarity in DNAR orders and the risk that they are interpreted or applied as instructions to limit treatment beyond resuscitation. Supplementary Information Additional file 1: ICD-10 codes for diagnoses and comorbidities. The ICD-10 codes were used to define the primary diagnoses and comorbidities. Additional file 2: Directed acyclic graph. A directed acyclic graph illustrated the assumed causal relationships between DNAR orders and discharge outcomes. DNAR orders were treated as the exposure and discharge outcome as the outcome. Age, activities of daily living (ADL), comorbidities, primary diagnoses, and severity on admission were considered potential confounders and were adjusted for in the regression models. Factors not captured in the structured database, including admission laboratory data, pre-admission DNAR status, prior advance care planning or palliative care consultation, management withdrawal after DNAR orders, and socioeconomic status, were considered potential unmeasured variables. Additional file 3: Supplementary Tables of Fine–Gray competing-risk analyses. This file contains Supplementary Table 2 and Supplementary Table 3 presenting results of Fine–Gray competing-risk regression analyses. Supplementary Table 2 shows the association between DNAR status and in-hospital death using a parsimonious model adjusted for age, sex, and Katz Index due to the limited number of events. Supplementary Table 3 presents multivariable analyses for discharge to home, adjusted for demographic characteristics, illness severity, primary diagnoses, and comorbidities. Additional file 4: Sensitivity analyses. Supplementary Table 4. After excluding these 62 patients and restricting the analysis to cases without missing covariates (n = 1,060), the association between DNAR status and transfer to post-acute care facilities remained similar to that observed in the primary analysis. Supplementary Table 5. When the Charlson Comorbidity Index was used instead of individual comorbidities, the results were unchanged. Supplementary Table 6. Analyses using more conservative confidence intervals based on a Bonferroni correction (98.33% confidence intervals) yielded essentially the same conclusions. Additional file 1: ICD-10 codes for diagnoses and comorbidities. The ICD-10 codes were used to define the primary diagnoses and comorbidities. Additional file 2: Directed acyclic graph. A directed acyclic graph illustrated the assumed causal relationships between DNAR orders and discharge outcomes. DNAR orders were treated as the exposure and discharge outcome as the outcome. Age, activities of daily living (ADL), comorbidities, primary diagnoses, and severity on admission were considered potential confounders and were adjusted for in the regression models. Factors not captured in the structured database, including admission laboratory data, pre-admission DNAR status, prior advance care planning or palliative care consultation, management withdrawal after DNAR orders, and socioeconomic status, were considered potential unmeasured variables. Additional file 3: Supplementary Tables of Fine–Gray competing-risk analyses. This file contains Supplementary Table 2 and Supplementary Table 3 presenting results of Fine–Gray competing-risk regression analyses. Supplementary Table 2 shows the association between DNAR status and in-hospital death using a parsimonious model adjusted for age, sex, and Katz Index due to the limited number of events. Supplementary Table 3 presents multivariable analyses for discharge to home, adjusted for demographic characteristics, illness severity, primary diagnoses, and comorbidities. Additional file 4: Sensitivity analyses. Supplementary Table 4. After excluding these 62 patients and restricting the analysis to cases without missing covariates (n = 1,060), the association between DNAR status and transfer to post-acute care facilities remained similar to that observed in the primary analysis. Supplementary Table 5. When the Charlson Comorbidity Index was used instead of individual comorbidities, the results were unchanged. Supplementary Table 6. Analyses using more conservative confidence intervals based on a Bonferroni correction (98.33% confidence intervals) yielded essentially the same conclusions.