Influencing factors analysis and prediction model development for ROSC in adult non-traumatic OHCA: a retrospective study based on pre-hospital emergency settings

Abstract

Background

Cardiac arrest is a major global public health problem. This study aimed to explore factors influencing return of spontaneous circulation (ROSC) and develop a prediction model for out-of-hospital cardiac arrest (OHCA) patients.

Methods

We retrospectively collected electronic medical records of adult non-traumatic OHCA patients at Tianjin Emergency Medical Center from January 1, 2023 to December 31, 2024. The primary outcome was sustained ROSC up to emergency department arrival and handover to receiving-hospital staff. Multivariable mixed-effects logistic regression with clinician ID as random intercept adjusted clustering bias; predictors with p < 0.05 formed the nomogram. Patients were randomly split 7:3 for training and internal testing. Discrimination was assessed by ROC-AUC (acceptable threshold >0.7). We performed 1000 bootstrap resampling to correct C-index against overfitting. Calibration curves and decision curve analysis (DCA) further evaluated predictive consistency and net clinical benefit.

Results

Among 9,386 included cases, 678 (7.2%) achieved ROSC. Mixed-effects regression identified independent ROSC predictors: public location (aOR = 1.285, 95% CI:1.038–1.592), 06:00–12:00 onset (aOR = 1.308, 95% CI:1.018–1.682), prolonged emergency response time (aOR = 0.971, 95% CI:0.945–0.999), extended onset-to-ambulance intervals (5–10 min: aOR = 0.758;10–30 min: aOR = 0.578;>30 min:aOR = 0.498 vs ≤5 min), no OHCA occurred at emergency call (aOR = 1.540, 95% CI:1.048–2.262), male attending physicians (aOR = 1.255, 95% CI:1.077–1.464), longer physician practice years (aOR = 1.043, 95%CI: 1.022–1.065), intermediate physician title (aOR = 1.236, 95% CI:1.047–1.459), shockable initial rhythm (aOR = 2.850, 95% CI:2.390–3.397), endotracheal intubation (aOR = 2.294, 95% CI:1.936–2.718), epinephrine >5 mg (aOR = 0.765,95% CI:0.626–0.935). The nomogram AUC reached 0.775 (95% CI:0.756–0.794) in the training set and 0.775 (95% CI:0.745–0.805) in the test set; 1000 bootstrap internal validation gave an optimism-corrected C-index of 0.7679 with minimal optimism of 0.0071, showing negligible overfitting, good calibration and favorable net clinical benefit via DCA.

Conclusion

This study identified multiple factors significantly associated with ROSC in OHCA patients; the constructed model showed good predictive value.

Introduction

Cardiac arrest is a life-threatening public health emergency with high mortality globally, posing enormous challenges to pre-hospital emergency medical services (EMS) and healthcare systems. A systematic review encompassing 133,981 OHCA cases treated by emergency medical services (EMS) across 33 countries globally revealed an incidence rate of 4.1 per 100,000 person-years, with significant continental disparities in incidence and survival outcomes (nearly a fourfold difference, from 16.2% in Australasia to 4.5% in Asia) [1]. In China, the burden of OHCA is particularly severe. In 2020, China reported an EMS-attended OHCA incidence rate of 95.7 per 100,000 population, and resuscitation was attempted for 31.8% patients [2]. A meta-analysis of the survival outcomes of out-of-hospital cardiac arrest patients in China after cardiopulmonary resuscitation showed that the rate of survival to hospital admission of only 5.0%, which was markedly lower than the global average of 22.0% [3]. In contrast, Europe and the United States annually treat approximately 275,000 and 155,000 adults with OHCA, respectively, with survival to hospital discharge rates ranging from 8% to 10% [4].

Following OHCA, cardiopulmonary resuscitation (CPR) or defibrillation are crucial intervention measures. Studies have shown that timely and effective CPR or defibrillation may restore spontaneous circulation (ROSC) in cardiac arrest patients. Otherwise, irreversible biological death will occur [5]. For China, clarifying the factors influencing ROSC in OHCA patients and establishing a reliable prediction model are essential to narrow the gap with developed countries and improve the effectiveness of pre-hospital OHCA resuscitation.

Different countries and different situations (such as differences in research design, sample size and demographic data, especially differences in hospital data and pre-hospital data of EMS) have different outcomes [6]. However, current studies in China on the factors influencing ROSC after OHCA are mainly small-sample studies based on in-hospital data. These studies fail to cover the entire pre-hospital emergency process, because in-hospital data only reflect post-admission management and cannot include key pre-hospital links that directly affect ROSC in OHCA patients (e.g. CPR performed by lay rescuers, EMS response time), which are directly related to the recovery of OHCA patients.

More importantly, few studies have utilized Chinese pre-hospital EMS data to systematically identify the factors influencing ROSC in OHCA patients, and no reliable ROSC prediction model based on large-sample pre-hospital emergency data has been established to guide clinical practice. This gap limits the ability to implement targeted pre-hospital interventions and optimize pre-hospital resuscitation outcomes for cardiac arrest patients in China.

Previous OHCA prognostic tools, including the RACA⁷ and UB-ROSC scores [7], have been widely validated in European population [8–10], yet external validation results in the population with a lower ROSC rate showed inconsistent predictive performance [11]. A recent Asian external validation study [12] further confirmed that RACA, UB-ROSC, and P-ROSC scores [13] all exhibit limited generalizability in real-world Asian pre-hospital OHCA cohorts, with no single tool achieving satisfactory and stable efficacy for regional populations. Such discrepancies are largely attributable to distinct EMS workflows, resuscitation protocols, and demographic characteristics between different communities. Notably, the newly proposed UB-ROSC score remains insufficiently validated in Chinese populations. Furthermore, most existing models are designed for the initial EMS assessment at dispatch/arrival, whereas few tools are dedicated to include the treatment information and patients’ situation during the ambulance transfer process, which is not conducive to the efficient connection between pre-hospital emergency care and in-hospital treatment.

Thus, this study had two core objectives: first, to systematically identify the key factors associated with ROSC in patients with OHCA using electronic medical record data from pre-hospital emergency care provided by the Tianjin Emergency Medical Center. Second, to develop and validate a reliable ROSC prediction model based on more complete OHCA-related data, which can be used during the ambulance transfer process and serve as a quantitative tool for retrospective prehospital resuscitation quality control.

Materials and methods

Data source

The data for this retrospective study were obtained from the Tianjin Emergency Medical Center’s dispatch communications and pre-hospital electronic medical records between January 1, 2023 and December 31, 2024. As the city’s sole municipal pre-hospital emergency medical service provider, the center serves 6 urban districts and 4 suburban districts, operating 109 emergency stations serving a resident population of approximately 8 million. It operates 290 ambulances equipped with critical care devices including defibrillator monitors, video laryngoscopes, and electrocardiographs. Each emergency team comprises a physician, nurse, driver, and two emergency assistants. ‘120’ is the nationwide emergency medical services hotline, directly connecting to regional emergency medical dispatch centers. Dispatchers dispatch ambulances upon receiving ‘120’ emergency calls, provide remote guidance to callers identified as cardiac arrest patients, record relevant time-stamped data, and utilize standardized medical report forms to collect demographic and clinical information.

Study population

This retrospective observational study enrolled adult patients with non-traumatic out-of-hospital cardiac arrest (OHCA) who received resuscitation attempts by the Tianjin Emergency Medical Center from January 1, 2023 to December 31, 2024. Cases were screened according to the following inclusion and exclusion criteria. Inclusion criteria: (1) Age ≥ 18 years; (2) Cardiac arrest occurred outside the hospital; (3) Received emergency treatment from the pre-hospital emergency rescue team (such as cardiopulmonary resuscitation or defibrillation). Exclusion criteria:(1) Age <18 years; (2) Cases with no resuscitation attempted or resuscitation explicitly refused by family (e.g. decapitated patients, severe brain/thoracic/abdominal organ destruction, obvious lividity/rigor mortis/putrefaction, explicit refusal of resuscitation by patient’s relatives); (3) Cardiac arrest that occurred within the hospital; (4) Traumatic cardiac arrest; (5) Incomplete records or missing data.

Data collection

This study collected the basic demographic information of the patients (sex, age) and OHCA time period (0:00–6:00, 6:00–12:00, 12:00–18:00, 18:00–24:00), season (spring, summer, autumn, winter), location (home, public location, nursing home), geographic area of cardiac arrest (urban vs. suburban), preexisting medical histories (cancer, cardiovascular disease, and respiratory disease), attending physician’s general information (sex, age, education level, professional title, years of practice), emergency response time (time from EMS call receipt to ambulance arrival), presence of shockable rhythm and administration of defibrillation, establishment of advanced airway via endotracheal intubation (ETI), epinephrine dosage(≤5mg, >5mg), use of mechanical cardiopulmonary resuscitation (mechanical CPR), time from onset to ambulance arrival, and lay rescuer CPR status was categorized into four groups: no CPR performed by lay rescuers, CPR performed by lay rescuers, no OHCA occurred at emergency call, and unrecognizable arrest onset status. All data were recorded in the dispatch communication system or electronic medical record system. Patients were categorized into ROSC and non-ROSC groups based on whether spontaneous circulation was restored and survival to hospital admission occurred.

OHCA and ROSC determination criteria

OHCA was defined as the loss of functional cardiac mechanical activity related to systemic circulation outside the hospital. The primary outcome was sustained ROSC until arrival at the emergency department and handover to receiving-hospital staff. Cases with brief pulse recovery during resuscitation or loss of circulation prior to hospital admission were uniformly classified into the non-ROSC group [14]. OHCA and ROSC were confirmed by the on-site emergency doctors and reviewed by the Medical Professional Technical Committee of Tianjin Emergency Medical Center. Clinical termination-of-resuscitation follows the Universal Termination-of-Resuscitation (TOR) criteria: field termination may be considered for non-EMS-witnessed arrest without ROSC and without defibrillation shocks; otherwise hospital transport is required. Even if TOR criteria are met, family-surrogate preferences are respected to balance standard protocols and ethical principles [15].

Statistical methods

For continuous variables, after normality testing, the normal distribution was expressed as the mean ± standard deviation. For non-normal distributed quantitative data, median (P₂₅,P₇₅) was used. Categorical variables were described using frequency (percentage). To compare inter-group differences, t-tests or nonparametric tests were used for quantitative data, while chi-square tests were applied for categorical data.

Variables with p < 0.1 in the univariate analysis were included in the multivariable mixed-effects logistic regression model to identify independent predictors of prehospital ROSC. Clinician ID was incorporated as a random intercept to account for clustering effect arising from the consistent treatment techniques of the same physician and to reduce statistical bias caused by correlated observations. Only variables with p < 0.05 in the final mixed-effects model were retained for nomogram construction.

After identifying independent predictors, all patients were randomly divided into a training set and an internal test set at a 7:3 ratio. An interpretable nomogram prediction model was constructed based on the statistically significant predictors derived from the mixed-effects logistic regression.

Receiver operating characteristic (ROC) curves and corresponding area under the curve (AUC) were calculated in both training and test sets to assess model discrimination. An AUC > 0.7 was regarded as acceptable discriminative ability. Internal validation via 1000 bootstrap resampling was implemented to quantify model optimism and correct the C-index to mitigate overfitting. Calibration curves were plotted to evaluate the consistency between model-predicted probabilities and actual observed ROSC rates. Decision curve analysis (DCA) was performed to quantify the net clinical benefit of the nomogram across different threshold probabilities and judge its clinical application value.

A two-sided p-value <0.05 was considered statistically significant. All statistical analyses were performed using R 4.3.0.

Results

Baseline characteristics of study subjects

From January 1, 2023 to December 31, 2024, a total of 18928 cases of OHCA admitted to Tianjin Emergency Medical Center were enrolled. Cases meeting the following exclusion criteria were excluded, including 121 cases involving patients under 18 years of age, 8,666 cases with no resuscitation attempted or resuscitation explicitly refused by family (e.g. decapitated patients, severe brain/thoracic/abdominal organ destruction, obvious lividity/rigor mortis/putrefaction, explicit refusal of resuscitation by patient’s relatives), 262 cases in the hospital where the occurrence occurred, 333 cases of traumatic cardiac arrest (Figure 1). All the above categories were outside the scope of this study. The additional 160 cases of missing key variable information were also excluded, of which the basic information was essentially consistent with the final included cases (Supplementary Table S1).

Figure 1.

Procedures for inclusion and exclusion of study subjects.

Among the 9,386 included OHCA cases, 678 (7.2%) were in the ROSC group and 8,708 (92.8%) in the non-ROSC group. The number of male patients was 6,387 (68.05%), while female patients numbered 2,999 (31.95%). The median age was 71.00 (62.00, 79.00) years. A comparative analysis of general characteristics between the ROSC group and non-ROSC group is presented in Table 1.

Table 1.

Basic characteristics of ROSC group and non-ROSC group.

Characteristic	Overall N = 9386	Non ROSCN = 8708	ROSC N = 678	P-value[1]
Arrest area				0.069
Suburban	3,911 (41.67%)	3,606 (41.41%)	305 (44.99%)
Urban	5,475 (58.33%)	5,102 (58.59%)	373 (55.01%)
Location, n (%)				<0.001
Home	7,786 (82.95%)	7,269 (83.47%)	517 (76.25%)
Public location	1,259 (13.41%)	1,124 (12.91%)	135 (19.91%)
Nursing home	341 (3.63%)	315 (3.62%)	26 (3.83%)
Season, n (%)				0.130
Spring	2,212 (23.57%)	2,049 (23.53%)	163 (24.04%)
Summer	2,128 (22.67%)	1,968 (22.60%)	160 (23.60%)
Autumn	2,391 (25.47%)	2,202 (25.29%)	189 (27.88%)
Winter	2,655 (28.29%)	2,489 (28.58%)	166 (24.48%)
Time period, n (%)				0.027
0:00–6:00	1,419 (15.12%)	1,343 (15.42%)	76 (11.21%)
6:00–12:00	3,002 (31.98%)	2,778 (31.90%)	224 (33.04%)
12:00–18:00	2,562 (27.30%)	2,373 (27.25%)	189 (27.88%)
18:00–24:00	2,403 (25.60%)	2,214 (25.42%)	189 (27.88%)
Time from onset to ambulance arrival, n (%)				<0.001
≤5 min	1,723 (18.36%)	1,526 (17.52%)	197 (29.06%)
5 ∼ 10 min	3,767 (40.13%)	3,441 (39.52%)	326 (48.08%)
10 ∼ 30 min	3,389 (36.11%)	3,243 (37.24%)	146 (21.53%)
>30 min	507 (5.40%)	498 (5.72%)	9 (1.33%)
Emergency response time, Median (Q1, Q3)	7.18 (5.68, 9.03)	7.20 (5.70, 9.05)	6.95 (5.40, 8.82)	0.002
Lay rescuer CPR status, n (%)				<0.001
No CPR performed by lay rescuers	726 (7.73%)	694 (7.97%)	32 (4.72%)
CPR performed by lay rescuers	4,356 (46.41%)	4,090 (46.97%)	266 (39.23%)
No OHCA occurred at emergency call	986 (10.51%)	865 (9.93%)	121 (17.85%)
Unrecognizable arrest onset status	3,318 (35.35%)	3,059 (35.13%)	259 (38.20%)
Physician age, Median, Mean (SD)	37.43 (4.90)	37.40 (4.90)	37.82 (4.95)	0.031
Physician sex, n (%)				<0.001
Female	4,386 (46.73%)	4,117 (47.28%)	269 (39.68%)
Male	5,000 (53.27%)	4,591 (52.72%)	409 (60.32%)
Physician practice years, Median (Q1, Q3)	4.54 (3.95, 5.15)	4.51 (3.93, 5.13)	4.85 (4.35, 5.35)	<0.001
Physician education, n (%)				0.035
Junior college degree	298 (3.17%)	285 (3.27%)	13 (1.92%)
Undergraduate college	8,042 (85.68%)	7,440 (85.44%)	602 (88.79%)
Graduate student	1,046 (11.14%)	983 (11.29%)	63 (9.29%)
Physician professional title, n (%)				<0.001
Junior	4,613 (49.15%)	4,337 (49.80%)	276 (40.71%)
Intermediate	4,440 (47.30%)	4,070 (46.74%)	370 (54.57%)
Senior	333 (3.55%)	301 (3.46%)	32 (4.72%)
First arresting rhythm, n (%)				<0.001
Unshockable rhythm	8,120 (86.51%)	7,688 (88.29%)	432 (63.72%)
Shockable rhythm	1,266 (13.49%)	1,020 (11.71%)	246 (36.28%)
Mechanical CPR				<0.001
No	7,561 (80.56%)	7,048 (80.94%)	513 (75.66%)
Yes	1,825 (19.44%)	1,660 (19.06%)	165 (24.34%)
Advanced airway by ETI, n (%)				<0.001
No	4,912 (52.33%)	4,742 (54.46%)	170 (25.07%)
Yes	4,474 (47.67%)	3,966 (45.54%)	508 (74.93%)
Epinephrine dose, n (%)				0.003
≤5 mg	7,756 (82.63%)	7,224 (82.96%)	532 (78.47%)
>5 mg	1,630 (17.37%)	1,484 (17.04%)	146 (21.53%)
Patient’s sex, n (%)				0.932
Female	2,999 (31.95%)	2,781 (31.94%)	218 (32.15%)
Male	6,387 (68.05%)	5,927 (68.06%)	460 (67.85%)
Patient’s age, Median (Q1, Q3)	71.00 (62.00, 79.00)	71.00 (62.00, 80.00)	69.50 (61.00, 77.00)	<0.001
Cancer, n (%)				0.943
No	8,454 (90.07%)	7,844 (90.08%)	610 (89.97%)
Yes	932 (9.93%)	864 (9.92%)	68 (10.03%)
Cardiovascular diseases, n (%)				0.962
No	6,936 (73.90%)	6,434 (73.89%)	502 (74.04%)
Yes	2,450 (26.10%)	2,274 (26.11%)	176 (25.96%)
Respiratory system diseases, n (%)				0.900
No	8,594 (91.56%)	7,972 (91.55%)	622 (91.74%)
Yes	792 (8.44%)	736 (8.45%)	56 (8.26%)

¹ Pearson’s Chi-squared test; Wilcoxon rank sum test; CPR, cardiopulmonary resuscitation; OHCA, out-of-hospital cardiac arrest; ETI, endotracheal intubation; ROSC, return of spontaneous circulation.

We observed statistically significant differences between the ROSC group and the non-ROSC group in OHCA location, OHCA time period, time from onset to ambulance arrival, emergency response time, lay rescuer CPR status, physician’s general information (age, sex, years of service, education level, professional title), first arresting rhythm, mechanical CPR, advanced airway established via ETI, epinephrine dosage, and patient age (p < 0.05). No statistically significant intergroup differences were detected for the remaining variables.

Multivariate analysis of influencing factors

A generalized linear mixed-effects logistic regression model was established in this study, with clinician ID specified as the random intercept to adjust for clustering effects arising from multiple patients managed by the same physician. The results is presented in Table 2.

Table 2.

Multivariate mixed-effects logistic regression analysis of ROSC rates following CPR among patients with non-traumatic OHCA.

Characteristic	β	Adjusted OR (95%CI)#	P-value
Arrest area
Suburban		1.000
Urban	−0.133	0.876 (0.745–1.030)	0.108
Location
Home		1.000
Public location	0.251	1.285 (1.038–1.592)	0.021
Nursing home	0.178	1.195 (0.802–1.782)	0.381
Time period
0:00–6:00		1.000
6:00–12:00	0.269	1.308 (1.018–1.682)	0.036
12:00–18:00	0.165	1.180 (0.913–1.525)	0.207
18:00–24:00	0.233	1.263 (0.977–1.631)	0.074
Patient’s age	0.001	1.001 (0.995–1.007)	0.710
Emergency response time	−0.029	0.971 (0.945–0.999)	0.040
Time from onset to ambulance arrival
≤5min		1.000
5 ∼ 10min	−0.277	0.758 (0.618–0.929)	0.007
10 ∼ 30min	−0.548	0.578 (0.458–0.730)	<0.001
>30min	−0.697	0.498 (0.308–0.730)	0.004
Lay rescuer CPR status, n (%)
No CPR performed by lay rescuers		1.000
CPR performed by lay rescuers	0.001	1.001 (0.717–1.397)	0.995
No OHCA occurred at emergency call	0.432	1.540 (1.048–2.262)	0.028
Unrecognizable arrest onset status	0.230	1.258 (0.900–1.759)	0.179
Physician sex
Female		1.000
Male	0.228	1.255 (1.077–1.464)	0.004
Physician age	−0.003	0.997 (0.979–1.014)	0.700
Physician practice years	0.042	1.043 (1.022–1.065)	<0.001
Physician education
Junior college degree		1.000
Undergraduate college	0.326	1.386 (0.836–2.299)	0.206
Graduate student	0.211	1.242 (0.705–2.162)	0.460
Physician professional title
Junior		1.000
Intermediate	0.212	1.236 (1.047–1.459)	0.012
Senior	0.280	1.324 (0.873–2.006)	0.186
Epinephrine dose
≤5mg		1.000
>5mg	−0.268	0.765 (0.626–0.935)	0.009
First arresting rhythm, n (%)
Unshockable rhythm		1.000
Shockable rhythm	1.047	2.850(2.390–3.397)	<0.001
Mechanical CPR
No		1.000
Yes	−0.013	0.987 (0.814–1.196)	0.893
Advanced airway by ETI
No		1.000
Yes	0.830	2.294(1.936–2.718)	<0.001

^# Results were derived from generalized linear mixed-effects logistic regression model, with a random intercept specified for physician ID to account for clustering of patients treated by the same physician. CPR, cardiopulmonary resuscitation; OHCA, out-of-hospital cardiac arrest; ETI, endotracheal intubation; ROSC, return of spontaneous circulation; OR, odds ratio; CI: confidence interval.

Compared to home settings, patients with cardiac arrest in public places had a higher probability of achieving ROSC (adjusted odds ratio [aOR] = 1.285, 95%CI: 1.038–1.592, p = 0.021). Cardiac arrest occurring during the 6:00–12:00 period was significantly associated with increased ROSC rates relative to the 00:00–06:00 timeframe (aOR = 1.308, 95%CI: 1.018–1.682, p = 0.036). Longer emergency response time was independently associated with a lower likelihood of ROSC (aOR = 0.971, 95%CI: 0.945–0.999, p = 0.040). Compared with an onset-to-ambulance arrival interval ≤5 min, prolonged intervals of 5–10 min (aOR = 0.758, 95%CI: 0.618–0.929, p = 0.007), 10–30 min (aOR = 0.578, 95%CI: 0.458–0.730, p < 0.001), and over 30 min (aOR = 0.498, 95%CI: 0.308–0.730, p = 0.004) were correlated with decreased ROSC probability. Cases without confirmed OHCA at emergency call presentation showed significantly higher ROSC likelihood (aOR = 1.540, 95%CI: 1.048–2.262, p = 0.028). No significant association was observed between lay rescuer CPR and ROSC.

In terms of physician-related factors, cases managed by male physicians exhibited higher ROSC probability (aOR = 1.255, 95%CI: 1.077–1.464, p = 0.004). Longer clinical practice years was positively associated with ROSC occurrence (aOR = 1.043, 95%CI: 1.022–1.065, p < 0.001). Relative to physicians with junior professional titles, intermediate-title physicians (aOR = 1.236, 95%CI: 1.047–1.459, p = 0.012) demonstrated significantly higher ROSC rates.

For resuscitation interventions, initial shockable rhythm was strongly associated with increased ROSC probability (aOR = 2.850, 95%CI: 2.390–3.397, p < 0.001). Patients receiving endotracheal intubation for advanced airway management had significantly higher ROSC likelihood (aOR = 2.294, 95%CI: 1.936–2.718, p < 0.001). In contrast, epinephrine dosage exceeding 5 mg was independently associated with reduced ROSC rates (aOR = 0.765, 95%CI: 0.626–0.935, p = 0.009). No statistically significant independent associations with ROSC were identified for the remaining candidate variables (all p > 0.05).

Multivariate binary logistic regression analysis was conducted to identify independent predictors of prehospital return of spontaneous circulation (ROSC). After screening, 11 variables with statistically significant independent associations with ROSC (p < 0.05) were retained to construct the predictive nomogram.

Construction of visual nomogram

All cases were randomly divided into a training set (N = 6571) and a test set (N = 2815) at a ratio of 7:3. A comparative analysis of general characteristics between the training and test sets is presented in Table 3, with no significant differences observed (p > 0.05).

Table 3.

Basic characteristics of training set and test set participants.

Characteristic	OverallN = 9386	Training setN = 6571	Test setN = 2815	P-value[1]
Arrest area				0.999
Suburban	3,911 (41.67%)	2,738 (41.67%)	1,173 (41.67%)
Urban	5,475 (58.33%)	3,833 (58.33%)	1,642 (58.33%)
Location				0.079
Home	7,786 (82.95%)	5,423 (82.53%)	2,363 (83.94%)
Public location	1,259 (13.41%)	892 (13.57%)	367 (13.04%)
Nursing home	341 (3.63%)	256 (3.90%)	85 (3.02%)
Season, n (%)				0.931
Spring	2,212 (23.57%)	1,548 (23.56%)	664 (23.59%)
Summer	2,128 (22.67%)	1,494 (22.74%)	634 (22.52%)
Autumn	2,391 (25.47%)	1,662 (25.29%)	729 (25.90%)
Winter	2,655 (28.29%)	1,867 (28.41%)	788 (27.99%)
Time period, n (%)				0.470
0:00–6:00	1,419 (15.12%)	1,004 (15.28%)	415 (14.74%)
6:00–12:00	3,002 (31.98%)	2,112 (32.14%)	890 (31.62%)
12:00–18:00	2,562 (27.30%)	1,803 (27.44%)	759 (26.96%)
18:00–24:00	2,403 (25.60%)	1,652 (25.14%)	751 (26.68%)
Time from onset to ambulance arrival, n (%)				0.637
≤5min	1,723 (18.36%)	1,201 (18.28%)	522 (18.54%)
5 ∼ 10min	3,767 (40.13%)	2,653 (40.37%)	1,114 (39.57%)
10 ∼ 30min	3,389 (36.11%)	2,353 (35.81%)	1,036 (36.80%)
>30min	507 (5.40%)	364 (5.54%)	143 (5.08%)
Emergency response time, Median (Q1, Q3)	7.80 ± 3.24	7.77 ± 3.24	7.85 ± 3.25	0.132
Lay rescuer CPR status, n (%)				0.620
No CPR performed by lay rescuers	726 (7.73%)	519 (7.90%)	207 (7.35%)
CPR performed by lay rescuers	4,356 (46.41%)	3,029 (46.10%)	1,327 (47.14%)
No OHCA occurred at emergency call	986 (10.51%)	701 (10.67%)	285 (10.12%)
Unrecognizable arrest onset status	3,318 (35.35%)	2,322 (35.34%)	996 (35.38%)
Physician age, Median (Q1, Q3)	37.19 (33.87, 41.04)	37.16 (33.87, 41.09)	37.27 (33.86, 40.96)	0.971
Physician sex, n (%)				0.156
Female	4,386 (46.73%)	3,102 (47.21%)	1,284 (45.61%)
Male	5,000 (53.27%)	3,469 (52.79%)	1,531 (54.39%)
Physician practice years, Median (Q1, Q3)	5.27 ± 3.14	5.23 ± 3.03	5.36 ± 3.39	0.874
Physician education, n (%)				0.931
Junior college degree	298 (3.17%)	206 (3.13%)	92 (3.27%)
Undergraduate college	8,042 (85.68%)	5,630 (85.68%)	2,412 (85.68%)
Graduate Student	1,046 (11.14%)	735 (11.19%)	311 (11.05%)
Physician professional title, n (%)				0.366
Junior	4,613 (49.15%)	3,245 (49.38%)	1,368 (48.60%)
Intermediate	4,440 (47.30%)	3,084 (46.93%)	1,356 (48.17%)
Senior	333 (3.55%)	242 (3.68%)	91 (3.23%)
First arresting rhythm, n (%)				0.172
Unshockable rhythm	8,120 (86.51%)	5,664 (86.20%)	2,456 (87.25%)
Shockable rhythm	1,266 (13.49%)	907 (13.80%)	359 (12.75%)
Mechanical CPR				0.404
No	7,561 (80.56%)	5,308 (80.78%)	2,253 (80.04%)
Yes	1,825 (19.44%)	1,263 (19.22%)	562 (19.96%)
Advanced airway by ETI, n (%)				0.448
No	4,912 (52.33%)	3,422 (52.08%)	1,490 (52.93%)
Yes	4,474 (47.67%)	3,149 (47.92%)	1,325 (47.07%)
Epinephrine dose, n (%)				0.760
≤5mg	7,756 (82.63%)	5,435 (82.71%)	2,321 (82.45%)
>5mg	1,630 (17.37%)	1,136 (17.29%)	494 (17.55%)
Patient’s sex, n (%)				0.544
Female	2,999 (31.95%)	2,087 (31.76%)	912 (32.40%)
Male	6,387 (68.05%)	4,484 (68.24%)	1,903 (67.60%)
Patient’s age, Median (Q1, Q3)	71.00 (62.00, 79.00)	71.00 (61.00, 79.00)	71.00 (62.00, 79.00)	0.341
ROSC				0.976
No	8,708 (92.78%)	6,096 (92.77%)	2,612 (92.79%)
Yes	678 (7.22%)	475 (7.23%)	203 (7.21%)
Cancer, n (%)				0.347
No	8,454 (90.07%)	5,931 (90.26%)	2,523 (89.63%)
Yes	932 (9.93%)	640 (9.74%)	292 (10.37%)
Cardiovascular diseases, n (%)				0.829
No	6,936 (73.90%)	4,860 (73.96%)	2,076 (73.75%)
Yes	2,450 (26.10%)	1,711 (26.04%)	739 (26.25%)
Respiratory system diseases, n (%)				0.658
No	8,594 (91.56%)	6,022 (91.65%)	2,572 (91.37%)
Yes	792 (8.44%)	549 (8.35%)	243 (8.63%)

¹ Pearson’s Chi-squared test; Wilcoxon rank sum test; CPR, cardiopulmonary resuscitation; OHCA, out-of-hospital cardiac arrest; ETI, endotracheal intubation; ROSC, return of spontaneous circulation.

A prediction model was constructed using 11 independent variables associated with ROSC to forecast the probability of ROSC occurrence in OHCA patients (Figure 2A,B). Each variable was assigned a score ranging from 0 to 100 based on its importance. We summed these scores to calculate a total score. This total score was then converted to a ROSC probability ranging from 0% to 100%. Additionally, to facilitate use and obtain more precise predictions, we developed a web-based calculator based on the scorecard (https://tianjin-emergency-medical-center.shinyapps.io/rosc-ohca-investigational-nomogram/). Notably, this web calculator has not undergone external validation and is intended solely as an investigational tool.

Figure 2.

Nomogram and web calculator. (A) A visual nomogram for predicting the ROSC after non-traumatic OHCA. (B) An example of a web calculator.

The logit linear predictor equation for calculating the log odds of ROSC is shown in Equation (1), and the formula for converting log odds to ROSC probability is presented in Equation (2).

$$\begin{array}{l}\text{Logit}\left(P\left(\text{ROSC}=1\right)\right)=-4.1070+0.2796X_{\text{PublicLocation}}-0.3362X_{\mathit{\text{Epinephrine}}>5\mathit{\text{mg}}}+0.3341X_{\mathit{\text{MalePhysician}}}+\\ 0.0483X_{\mathit{\text{PracticeYears}}}+0.3105X_{\mathit{\text{IntermediateTitle}}}+1.1527X_{\mathit{\text{ShockableRhythm}}}+1.2297X_{\mathit{\text{ETI}}}-0.4134X_{\mathit{\text{Arrest}}5-10\mathit{\text{min}}}-\\ 0.8377X_{\mathit{\text{Arrest}}10-30\mathit{\text{min}}}-1.2141X_{\mathit{\text{Arrest}}>30\mathit{\text{min}}}+0.7038X_{\mathit{\text{Post}}-\mathit{\text{callOHCA}}}\end{array}$$ (1)

Calculation of ROSC probability:

$$P(\text{ROSC}\mathrm{)}\text{=}\frac{\mathrm{\text{exp}}\left[\mathit{\text{Logit}}\mathrm{(}P\right(\mathit{\text{ROSC}}=1\mathrm{))}\text{]}}{\text{1+exp[Logit}\mathrm{(}P(\mathit{\text{ROSC}}=1\mathrm{))}\text{]}}$$ (2)

For example, a patient experiencing out-of-hospital cardiac arrest (OHCA) in a public place between 18:00 and 24:00, with bystander CPR performed by lay rescuers, a 5–10 min interval from cardiac arrest onset to ambulance arrival, and a 10-min EMS response time. Resuscitation was delivered by a male emergency physician with 15 years of clinical practice and an intermediate professional title; the patient presented an initial shockable rhythm, received epinephrine at a dose >5 mg, and underwent endotracheal intubation. The nomogram predicted a 27.2% probability of return of spontaneous circulation (ROSC) for this patient (Figure 2).

It should be noted that only standardized clinical variables including age, sex and resuscitation interventions can be entered, with no input boxes for identifiable personal information such as names and ID numbers to protect patient privacy. The calculator will undergo iterative updates of model parameters according to subsequent external validation findings and updated emergency guidelines. Notably, this tool was developed from retrospective data, so large-scale clinical application is not recommended currently.

Performance of the nomogram

The multivariable logistic regression prediction model developed in this study exhibited favourable discrimination, satisfactory calibration, and meaningful clinical net benefit for patients with out-of-hospital cardiac arrest (OHCA). The AUC was 0.775 (95% CI: 0.756–0.794) in the training cohort and 0.775 (95% CI: 0.745–0.805) in the independent validation cohort. Internal validation via 1000 bootstrap resamples yielded an optimism-corrected C-index of 0.7679 with a minimal optimism of 0.0071, indicating negligible overfitting. At the optimal probability threshold of 0.065 derived from the training cohort, the model yielded balanced sensitivity of 78.1%, specificity of 65.7%, positive predictive value (PPV) of 15.1%, negative predictive value (NPV) of 97.5%, and an overall accuracy of 66.6% (Figure 3). Comprehensive calibration assessments yielded consistent favourable results: the independent validation cohort had a calibration intercept of 0.099 and slope of 1.0361; the apparent Brier score was 0.0627 for the training set, 0.0622 for the validation set, and 0.0631 after bootstrap correction. Calibration plots demonstrated strong concordance between predicted (Figure 4A,B) and observed probabilities, and the validation cohort passed the Hosmer–Lemeshow test (p = 0.077). Decision curve analysis confirmed that the prediction model provided higher net benefit than the two reference strategies (treating all patients or treating none) within the threshold probability range of 0 ∼ 0.22 (Figure 5A,B). Collectively, the model achieved stable discrimination, minimal predictive bias, limited overfitting, and possesses practical clinical utility for individualized prehospital ROSC risk prediction.

Figure 3.

ROC curve of the predicting model of the training set and test set. The optimal probability cutoff value (0.065) and corresponding diagnostic metrics, including sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), Brier score, and accuracy, are annotated as text within the plot. The 95% confidence intervals for the area under the curve (AUC) are provided in parentheses. AUC, area under the curve; ROC, receiver operating characteristic.

Figure 4.

Calibration plot of the predicting model. Panel (A) displays the training cohort, including the apparent curve and bias‑corrected curve derived from 1000‑bootstrap resamples. Panel (B) shows results for the independent test cohort. The 45° dashed ideal line represents perfect predictive accuracy. The x‑axis indicates predicted probability and the y‑axis denotes observed probability.

Figure 5.

Decision curve of the predicting model.

Discussion

The service scope of the Tianjin Emergency Medical Center covers 10 administrative districts in Tianjin, China (with a population of 8 million). Study data were obtained from the electronic medical records of the center between 2023 and 2024, including a total of 9386 out-of-hospital cardiac arrest (OHCA) cases, with an overall return of spontaneous circulation (ROSC) rate of 7.2%. This study first identified factors associated with ROSC in adult non-traumatic OHCA patients. Patient-related, event-related, emergency medical service-related, and treatment-related factors were significantly associated with ROSC outcome. A ROSC prediction model and a web-based calculator were then developed. It can be used during the ambulance transfer process and serve as a quantitative tool for retrospective prehospital resuscitation quality control. The multi-parameter model built using these associated variables showed good predictive performance and clinical utility for estimating ROSC in OHCA patients. By comparison, endpoints like ‘hospital discharge’ and ‘1-year survival’ are more clinically relevant but are affected by non-standardized post-resuscitation care, making them unsuitable for studies based primarily on prehospital emergency data.

Impact of OHCA patients and social factors on ROSC and public health implications

ROSC incidence also exhibits circadian variations. Patients with OHCA occurring during daytime or evening (excluding midnight to 6:00 AM) presented higher ROSC rates. Multiple prior studies observed increased mortality and reduced 1-month survival among patients experiencing nighttime OHCA, a phenomenon hypothesized to correlate with circadian activity patterns. OHCA events arising in daytime or evening hours tend to be recognized sooner and more frequently receive bystander CPR, which may correlate with improved resuscitation outcomes [16–18].

Our study showed higher ROSC rates in OHCA patients occurring in public settings, due to higher witness rates and faster emergency care access. In our study, lay rescuer CPR implementation reached 46.41%. Although no significant association between prehospital lay rescuer CPR prior to ambulance arrival and ROSC was observed in the present study, this lack of association may be explained by suboptimal CPR quality and extremely low rates of layperson AED use. Widespread public training in CPR and AED deployment may still help improve resuscitation outcomes. We also found that initial shockable rhythm and subsequent defibrillation were associated with higher ROSC rates. Shockable rhythms (e.g. ventricular fibrillation) usually present early in cardiac arrest with preserved organ perfusion and high resuscitation potential, whereas non-shockable rhythms reflect severe cardiac dysfunction and lower success rates. Early defibrillation thus improves ROSC rates [19,20]. Previous studies confirmed higher hospital discharge survival in public-setting OHCA patients treated with lay-rescuer AEDs, as these patients are often younger. Conversely, home-occurring OHCA more frequently involves elderly patients with complex comorbidities or end-stage disease, leading to a higher proportion of non-shockable rhythms [21,22].

Association between prehospital interventions and ROSC in OHCA patients and directions for clinical practice optimization

Time is critical for cardiac arrest patients. Irreversible brain injury typically occurs within 4–6 min, leading to biological death shortly thereafter [10]. Emergency response time was defined as the interval from the 120 emergency call receipt to ambulance arrival. We observed a 3.9% lower ROSC rate associated with each 1-minute longer emergency response interval. The median response time in our study was 7.18 min, meeting the widely recognized gold standard of 8 min [23,24].

Prior studies confirm that dispatchers’ early recognition of OHCA and telephone-guided bystander CPR facilitate timely resuscitation [25]. However, our study detected no significant independent association between layperson CPR before ambulance arrival and ROSC. This possibly due to inconsistent lay rescuer cooperation, variable CPR quality and extremely low public AED access. We found shockable-rhythm defibrillation (aOR = 2.850) and ETI (aOR = 2.294) were strong independent predictors of sustained ROSC. As a standard advanced airway strategy, ETI secures ventilation, optimizes oxygenation and lowers aspiration risk, while boosting oxygen delivery to preserve cardiac and cerebral function and promote ROSC [20,26]. Benoit et al. [27] similarly reported that early ETI improves ROSC likelihood. However, ETI has limitations: inexperienced providers face low success rates, and intubation pauses chest compressions to worsen prognosis; video laryngoscopy can minimize compression interruptions and raise first-pass success [28].

Our analysis identified cumulative epinephrine above 5 mg was independently negatively associated with ROSC. Epinephrine boosts left ventricular and cerebral perfusion via α-1 activation and elevated aortic diastolic pressure to facilitate ROSC during compressions [29]. However, high-dose resuscitation administration correlates with poorer post-arrest neurological outcomes [30,31]. A U.S. study reported higher survival in patients receiving epinephrine less frequently than the recommended 3–5-minute interval, indirectly suggesting harm from high doses [32]. Separate research reported an independent association between cumulative epinephrine >3 mg and failure to attain ROSC versus doses <3 mg (aOR = 3.98, 95%CI 2.43–6.51), alongside concurrent endocardial vasoconstriction and higher rates of malignant arrhythmia [20]. Since ERC guidelines advise epinephrine every 3–5 min [33]. Patients unresponsive to initial therapy receive higher doses, with >5 mg typically associated with >15 min of resuscitation. As such, the unfavorable relationship between total epinephrine dose and outcome can be partially attributed to ‘resuscitation time bias’ and underlying comorbidities [34].

The role of emergency personnel competency and teamwork in achieving ROSC in OHCA patients

A systematic review found a consistent positive association between EMS practitioners’ professional experience and improved OHCA survival [35]. Our findings corroborate this: senior physician titles are correlated with greater clinical competence, and longer service years coincide with greater cumulative exposure to OHCA cases alongside higher resuscitation success rates. Therefore, targeted professional training for less experienced, professional titles physicians is needed to standardize competence and treatment quality, reducing disparities and enhancing resuscitation skills and confidence [36].

We also found that patients treated by male emergency physicians had a 25.5% higher probability of achieving ROSC. This correlation must be interpreted with caution, of which the reasons might include the following aspects. Multiple confounding operational variables, such as prehospital crew composition and shift scheduling, may largely account for the observed disparity. High-quality CPR—adequate compression depth/rate, full chest recoil, minimal interruptions, and no hyperventilation—is critical for resuscitation and survival [37]. Prior simulation studies have demonstrated that CPR performance gradually deteriorates among all rescuers as resuscitation efforts are prolonged. Inter-individual differences in endurance are primarily driven by physical fatigue and team rotation protocols rather than provider gender [38]. Simulation studies also indicate substantial inter-rescuer variability in CPR quality, with lighter or smaller-statured rescuers more prone to fatigue [39]. This may explain the correlation that the higher resuscitation success was observed in male providers.

Prediction models for OHCA-ROSC

To our knowledge, this is the first study in China based on large-scale medical record data from an independent prehospital emergency medical institution, which comprehensively considers patient-related factors, event-related factors, emergency medical service-related factors, and treatment measure-related factors. A series of international models including RACA [40], UB-ROSC [7], WATCH-CPR [6], PRAD-CCPR [41] and P-ROSC [13] have been developed to predict ROSC in OHCA patients. Variations in sample sizes, outcome definitions and selected predictors lead to variable predictive performance. Though widely applied in European populations, cross-regional external validations have demonstrated inconsistent discrimination when these tools are transferred to Asian cohorts.

The German RACA model (2011) [40] and European UB-ROSC model [7] both adopted basic demographic and scene predictors such as age, arrest location, bystander CPR and initial rhythm, with AUCs of 0.731 and 0.83 respectively. However, UB-ROSC data were collected during the COVID-19 pandemic without thorough assessment of pandemic confounding effects. Two Thai models, WATCH-CPR [6] and PRAD-CCPR [41], are limited by small sample sizes, inadequate validation, missing core predictors and notable selection bias, such as exclusion of bystander-related variables. The 2022 P-ROSC score [13] incorporated epinephrine dosage and witnessed arrest and achieved an AUC of 0.806. A recent Asian validation study by Fan et al. [12] further verified that European scoring systems including RACA, UB-ROSC and P-ROSC exhibit unstable predictive efficacy in Taiwanese OHCA patients. Divergent population characteristics, EMS workflows and resuscitation protocols between European and Asian regions impair cross-cultural generalizability, highlighting an unmet need for localized prognostic tools suitable for mainland China’s prehospital emergency system.

Moreover, most existing models are designed for the initial EMS assessment at dispatch/arrival, whereas few tools are dedicated to include the treatment information and patients’ situation during the ambulance transfer process, which is not conducive to the efficient connection between pre-hospital emergency care and in-hospital treatment. Our study enrolled 9,386 OHCA cases and constructed a predictive model with five categories of predictors: patient demographics, arrest features, EMS response indicators, rescuer characteristics and full-spectrum prehospital interventions. Rigorous internal validation yielded an AUC of 0.775 in both training and test sets. This model boasts better representativeness and clinical suitability for mainland Chinese prehospital populations and fills the gap of localized OHCA-ROSC prediction raised by Fan et al. Notably, a low model-predicted ROSC probability never justifies stopping CPR, and resuscitation should be attempted for all OHCA patients.

Strengths and limitations

A notable strength of this study lies in exploring the influencing factors of ROSC in OHCA patients and constructing a prediction model based on ten administrative districts in Tianjin, China, covering a population of 8 million. This model comprehensively incorporates OHCA-related factors, pre-arrival interventions by emergency medical services (EMS), prehospital emergency treatment methods, and physician characteristics. The sample size is sufficient and has strong representativeness. Nevertheless, several limitations should be acknowledged. This retrospective analysis relied on archived medical records. Key information such as arrest witness profiles, detailed bystander characteristics, objective CPR quality metrics and time-stamped timelines for resuscitation interventions was not routinely documented and difficult to retrieve. We plan to standardize Utstein-style data collection in future prospective studies to address this gap. In addition, the model was constructed solely based on prehospital data without integrating key post-hospitalization interventions. Therefore, its application is more suitable for early risk identification and preliminary prognosis assessment in OHCA, making it difficult to provide definitive judgments on patients’ neurological recovery and long-term outcomes. Future studies will optimize prehospital information collection protocols, enable it to cover more comprehensive factors, and collaborate with hospitals to integrate post-hospitalization intervention data and long-term follow-up results, thereby improving the model’s applicability in predicting patients’ long-term prognosis and neurological recovery.

Conclusion

This study found that among non-traumatic OHCA patients in ten administrative districts of Tianjin during 2023–2024, the incidence of ROSC was 7.2%. The Multiple independent predictors were screened via multivariable mixed-effects logistic regression, and the established nomogram exhibited reliable predictive performance. The web-based calculator supports post-hoc prehospital emergency quality evaluation. This work carries practical value for optimizing regional public health emergency responses, promoting bystander rescue awareness, and lifting overall resuscitation success rates of cardiac arrest.

Funding Statement

This work was sponsored by Tianjin Health Research Project (grant NO. TJWJ2023MS043).

Ethics approval

This research has been approved by the Medical Ethics Committee of Tianjin Emergency Medical Center (approval NO. TJEMC-2025-003). All procedures involving human subjects were performed in accordance with the ethical principles of the World Medical Association Declaration of Helsinki. All patient data in this study were de-identified to eliminate all identifiable private information. Since this is a retrospective study, patient informed consent was waived.

Disclosure statement

We declare no competing interest.

Data availability statement

The raw data used in this study cannot be made publicly available due to privacy concerns, but are available from the corresponding author on reasonable request.