Factors associated with occult acute rib fractures on chest CT after negative radiography in older fall patients

Abstract

In older patients with fall-related injuries, chest computed tomography (CT) can detect acute rib fractures after chest radiography (CXR) showed no evidence of acute fracture. We evaluated clinical factors associated with CT-reported acute rib fracture within this clinician-selected imaging pathway. In this single-center retrospective observational study, we included patients aged ≥65 years who presented to the emergency department after a slip or fall between February 2021 and January 2024 and underwent same-visit chest CT after initial CXR without a definite, possible, or suspected acute rib fracture. The primary multivariable logistic model included age category (65–74, 75–84, or ≥85 years), male sex, fall from an elevated surface, non-alert mental status, head trauma, and localized chest wall pain. Covariates were entered simultaneously without automated selection, and all 739 patients were included. Sensitivity analyses examined alternative age specifications, sparse focal findings, and selectively ordered laboratory testing. Among 739 patients, 224 (30.3%) had a CT-reported acute rib fracture. Relative to ages 65 to 74 years, adjusted odds ratios (aORs) were 2.236 (95% confidence interval [CI], 1.443–3.465) for ages 75 to 84 years and 8.391 (95% CI, 4.354–16.172) for ages ≥85 years. Male sex (aOR, 1.791; 95% CI, 1.187–2.701), fall from an elevated surface (aOR, 2.415; 95% CI, 1.561–3.734), non-alert mental status (aOR, 1.891; 95% CI, 1.017–3.514), and localized chest wall pain (aOR, 11.541; 95% CI, 7.584–17.562) were also associated with acute rib fracture. Head trauma was not significantly associated with acute rib fracture. A 3-knot restricted cubic spline sensitivity analysis showed no evidence of departure from a linear age effect (P = .875); the estimate for non-alert mental status was attenuated when age was modeled continuously (P = .068). Within this clinician-selected CT pathway, older age, male sex, fall from an elevated surface, and localized chest wall pain were associated with CT-reported acute rib fracture after adjustment in the primary full-cohort model; the association with non-alert mental status was less stable across age specifications. These findings do not define a CT decision rule or support routine CT after negative CXR.

1. Introduction

Falls are an increasingly common reason for emergency department (ED) visits among older adults.^[1–3] Thoracic injury can be difficult to assess in this population because symptoms may be nonspecific, communication or examination may be limited, and subtle fractures may be missed on initial chest radiography (CXR).^[4,5] Rib fractures are clinically important because pain-related hypoventilation, atelectasis, and pneumonia contribute to adverse outcomes, and increasing fracture burden has been associated with greater morbidity and mortality.^[6–8]

CXR is accessible and commonly used for initial assessment but has limited sensitivity for rib fractures and associated thoracic injuries.^[5,9,10] Computed tomography (CT) detects more fractures and other thoracic injuries; however, CT-only findings do not consistently alter procedures, mortality, or other outcomes.^[9–12] Although some occult fractures may influence admission or monitoring, routine CT after every negative CXR may be inefficient and is associated with additional radiation, cost, and resource use.^[12–14]

Previous studies have largely examined broader blunt-trauma populations or focused on the diagnostic yield, resource use, and outcomes associated with CT-only fractures.^[9–12] Evidence remains limited on presentation-level characteristics among older fall patients who underwent same-visit CT after a CXR report without a definite, possible, or suspected acute fracture. Moreover, presentation can be atypical in older patients.^[15] We therefore evaluated clinical factors associated with CT-reported acute rib fracture within this imaging pathway.

2. Methods

2.1. Study design and population

This single-center retrospective observational study was conducted in the ED of a 600-bed university hospital that evaluates approximately 35,000 patients annually and serves as a regional Level I trauma center. Using a study-screening query, we identified 1387 potentially eligible ED injury encounters between February 1, 2021, and January 30, 2024, involving patients aged ≥65 years. The extract was created for study screening and was not an institutional census of all older fall-related visits.

After sequential review, we excluded 299 interhospital transfers, 78 encounters with mechanisms other than slipping or falling, 21 encounters with a definite, possible, or suspected acute rib fracture on initial CXR, 232 encounters without chest CT during the same index ED visit, and 18 duplicate non-index repeat visits. Only the first visit per patient was retained. No a priori sample-size calculation was performed; all eligible unique index visits in the source extract were included.

The final cohort comprised 739 unique patients: 224 with a CT-reported acute rib fracture and 515 with no CT-reported acute rib fracture (Fig. 1). The Institutional Review Board of Daejeon Eulji University Hospital approved the study and waived the requirement for individual informed consent because of its retrospective design (EMC 2025-11-004).

Figure 1.

Flow diagram of patient selection. Exclusions were applied sequentially. CT = computed tomography, CXR = chest radiography, ED = emergency department.

No formal institutional protocol mandated CT after negative CXR. Treating emergency physicians and the trauma team ordered CT according to clinical judgment, including consideration of the injury mechanism, focal thoracic findings, physiologic status or oxygen requirements, examination reliability, suspected associated injuries, and the need for broader trauma assessment. Structured patient-level ordering indications were unavailable, and CT was not necessarily obtained solely to detect rib fractures.

2.2. Data collection

Collected variables included age, sex, and injury mechanism; Korean Triage and Acuity Scale (KTAS) level, Glasgow Coma Scale score, alertness, vital signs, and oxygen therapy; arterial blood gas (ABG) measures (pH, partial pressure of oxygen, and lactate); white blood cell count, hemoglobin, C-reactive protein, renal and hepatic measures, calcium, phosphorus, electrolytes, and albumin; smoking status, comorbidities, and preinjury medications; and trauma-related findings, including head trauma, dyspnea, and location-specific pain. Comorbidities and medications were selected to capture recognized geriatric fall-risk and trauma-related factors.^[16–26]

Non-alert mental status denoted any documented mental-status category other than alert and was not derived from the Glasgow Coma Scale score.

Injury mechanism was classified from the documented event description as a ground-level slip or trip versus a fall from a bed, chair, stairs, or another elevated surface. A numerical fall-height field in the database was excluded from analysis and not reported quantitatively because its physical unit and derivation rule could not be verified. Age was grouped as 65 to 74, 75 to 84, and ≥85 years for the primary analysis; continuous and spline specifications were examined in sensitivity analyses.^[27]

The first author performed the primary chart abstraction using a standardized form. Ambiguous records were jointly reviewed by the first and corresponding authors and resolved by consensus. Independent duplicate abstraction and inter-rater reliability testing were not performed. Pain and dyspnea variables were based solely on the initial physician and nursing assessments; affirmative documentation defined their presence. The analytic dataset contained no separate missing category for most symptom fields; therefore, records without affirmative documentation, including those with unassessable symptoms, were coded absent. Localized sternal pain and abdominal pain were exceptions: 1 sternal pain value was unrecorded in the acute-fracture group, and 4 abdominal pain values were explicitly recorded as missing in the no-acute-fracture group. For the rare-pain sensitivity analysis, the single unrecorded sternal pain value was coded absent under this coding convention. This approach may have caused under-ascertainment or misclassification. Post-CT documentation was not used to redefine initial symptoms. Comorbidities and preinjury medications were identified from available records; statin use had one explicitly missing value in the no-acute-fracture group, whereas other medication fields without documentation were coded absent.^[28]

Initial imaging was generally a single frontal CXR, most often obtained in the anteroposterior view; dedicated rib series and routine lateral views were not standard. CXR was classified as negative only when the final report did not identify or raise suspicion of an acute rib fracture. Definite, possible, or suspected fractures were excluded. Nonspecific wording about osseous findings or chronic deformity without a definite, possible, or suspected acute fracture was not classified as positive.

Qualifying CT examinations were noncontrast or contrast-enhanced, covered the entire bilateral thorax, used 2- or 3-mm axial sections, and included coronal and sagittal reconstructions. Bone and soft-tissue window settings were available in the picture archiving and communication system. Acute fracture required explicit characterization as acute in the final CT report; reports of no fracture or of subacute, chronic, healed, or old fractures were classified as no CT-reported acute rib fracture. Three board-certified radiologists with more than seven years of experience interpreted the CXRs, and 1 thoracic radiologist issued the CT reports. Radiologists were not blinded to clinical information or prior imaging; investigators did not reinterpret the images, and interobserver agreement was not assessed.

2.3. Statistical analysis

Descriptive statistics and group comparisons in Tables 1 and 2 were performed using Statistical Package for the Social Sciences Statistics version 22.0 (IBM Corp., Armonk). The univariable and multivariable logistic regression analyses for the variables displayed in Table 3, the group comparison in Table S2, Supplemental Digital Content 2, and all sensitivity analyses were performed in Python version 3.11.15 (Python Software Foundation, Wilmington) using NumPy version 2.4.4 (NumPy Developers/NumFOCUS, Inc., Austin), SciPy version 1.17.1 (scipy.optimize and scipy.stats modules; SciPy Developers/NumFOCUS, Inc., Austin), and pandas version 3.0.2 (pandas development team/NumFOCUS, Inc., Austin); no additional statistical package was used. Standard logistic regression was fitted by maximum likelihood using Newton–Raphson iteration, with Wald 95% confidence intervals (CIs) and Wald P values. Firth logistic regression was implemented from the modified score equations of the Jeffreys-prior penalized likelihood,^[29,30] with profile penalized-likelihood CIs obtained by numerical root-finding on the penalized likelihood-ratio statistic and penalized likelihood-ratio P values. Restricted cubic splines used the truncated-power parameterization with linear-tail constraints,^[31] and departure from linearity was tested with a 1-df likelihood-ratio test. All routines are contained in a single documented script that also includes a validation suite executed before each analysis run: the maximum-likelihood routine reproduced published reference estimates and standard errors for the Bliss beetle-mortality data,^[32] the Firth routine reproduced the closed-form add-one-half solution for a saturated 2 × 2 table,^[29] yielded finite estimates under complete separation, and agreed with an independent numerical maximization of the penalized likelihood, and the spline basis satisfied the linear-tail constraints. The complete analysis code, package versions, and validation output are provided as Supplemental Digital Content S4, Supplemental Digital Content 1. In Statistical Package for the Social Sciences, normality was assessed with the Kolmogorov–Smirnov test. Continuous variables were compared using the Student’s t test and reported as mean ± standard deviation when normally distributed, or using the Mann–Whitney U test and reported as median [interquartile range] otherwise. Categorical variables were compared using Pearson chi-square tests, with continuity correction for 2 × 2 tables, and were reported as number (%). We did not adjust for multiple comparisons; unadjusted comparisons were considered exploratory.

Table 1.

Baseline characteristics according to acute rib fracture on chest computed tomography.

Variable	No CT-reported acute rib fracture (n = 515)	CT-reported acute rib fracture (n = 224)	P value
Demographics and mechanism of injury
Age (yr)	73.0 [66.0–79.0]	76.0 [70.0–82.0]	<.001
Age group			<.001
65–74 yr	306 (59.4)	100 (44.6)
75–84 yr	176 (34.2)	88 (39.3)
≥85 yr	33 (6.4)	36 (16.1)
Sex			<.001
Female	331 (64.3)	101 (45.1)
Male	184 (35.7)	123 (54.9)
Mechanism of injury			<.001
Ground-level slip/trip	359 (69.7)	107 (47.8)
Fall from an elevated surface	156 (30.3)	117 (52.2)
Emergency department severity and vital signs
Korean Triage and Acuity Scale	4.0 [3.0–4.0]	3.0 [3.0–4.0]	<.001
Glasgow Coma Scale score*	15.0 [15.0–15.0]	15.0 [15.0–15.0]	.270
Mental status			.003
Alert	477 (92.6)	193 (86.2)
Verbal response	15 (2.9)	20 (8.9)
Pain response	17 (3.3)	10 (4.5)
Unresponsive	6 (1.2)	1 (0.4)
Mental status, dichotomized			.008
Alert	477 (92.6)	193 (86.2)
Not alert	38 (7.4)	31 (13.8)
Systolic blood pressure (mm Hg)†	147.0 [130.0–164.0]	146.0 [122.0–167.0]	.448
Diastolic blood pressure (mm Hg)†	82.0 [73.0–91.0]	82.0 [74.0–92.0]	.680
Mean arterial pressure (mm Hg)†	103.3 [93.5–115.2]	103.5 [92.0–115.0]	.911
Heart rate (beats/min)†	80.0 [70.0–91.0]	83.0 [72.0–94.0]	.057
Respiratory rate (breaths/min)‡	20.0 [18.0–20.0]	20.0 [18.0–20.0]	.821
Body temperature (°C)†	36.7 [36.3–36.9]	36.6 [36.1–36.9]	.003
Peripheral oxygen saturation (%)§	98.0 [96.0–99.0]	97.0 [96.0–99.0]	.448
Oxygen therapy	110 (21.4)	94 (42.0)	<.001
Past medical history
Hypertension	249 (48.3)	108 (48.2)	>.999
Diabetes mellitus	142 (27.6)	58 (25.9)	.702
Tuberculosis	4 (0.8)	0 (0.0)	.437
Hepatitis	1 (0.2)	0 (0.0)	>.999
Hyperlipidemia	71 (13.8)	27 (12.1)	.603
Cardiovascular disease	33 (6.4)	15 (6.7)	>.999
Pulmonary disease	100 (19.4)	41 (18.3)	.801
Smoking, medications, and risk-related factors
Current smoking	55 (10.7)	40 (17.9)	.010
Hypnotics	80 (15.5)	42 (18.8)	.330
Antidepressants	60 (11.7)	33 (14.7)	.298
Anticholinergics	37 (7.2)	14 (6.2)	.762
Steroids	13 (2.5)	4 (1.8)	.727
Antiepileptic drugs	6 (1.2)	2 (0.9)	>.999
Thyroid hormones	20 (3.9)	4 (1.8)	.210
Thiazolidinediones	7 (1.4)	4 (1.8)	.913
Hormones	4 (0.8)	0 (0.0)	.437
Anticoagulants	34 (6.6)	16 (7.1)	.913
Antiplatelet agents	142 (27.6)	48 (21.4)	.096
Statins‖	155 (30.2)	62 (27.7)	.554
Trauma-related variables and localized symptoms
Dyspnea	66 (12.8)	43 (19.2)	.033
Head trauma	173 (33.6)	110 (49.1)	<.001
Localized neck pain	30 (5.8)	18 (8.0)	.338
Localized chest wall pain	60 (11.7)	128 (57.1)	<.001
Localized sternal pain¶	8 (1.6)	11 (4.9)	.016
Localized clavicular pain	12 (2.3)	29 (12.9)	<.001
Localized shoulder pain	72 (14.0)	48 (21.4)	.016
Localized arm pain	87 (16.9)	35 (15.6)	.750
Localized lower back pain (lumbar spine)	108 (21.0)	26 (11.6)	.003
Localized thoracic back pain (thoracic spine)	82 (15.9)	38 (17.0)	.807
Localized flank pain	45 (8.7)	20 (8.9)	>.999
Localized hip pain	119 (23.1)	38 (17.0)	.075
Abdominal pain#	54 (10.6)	18 (8.0)	.353

Data are median [interquartile range] or number (%).

All other displayed variables were available for the full group denominators.

Except for values explicitly recorded as missing, symptom fields had no separate missing category in the analytic dataset: affirmative documentation defined presence, and other records were coded absent, which may underestimate symptom prevalence.

CT = computed tomography.

^*Glasgow Coma Scale score, 493/207.

^†Systolic and diastolic blood pressure, mean arterial pressure, heart rate, and body temperature, 515/222.

^‡Respiratory rate, 513/222.

^§Peripheral oxygen saturation, 140/116.

^‖Statin use, 514/224.

^¶Localized sternal pain, 515/223.

^#Abdominal pain, 511/224.

Table 2.

Arterial blood gas and laboratory findings according to acute rib fracture on chest computed tomography.

Variable	No CT-reported acute rib fracture (n = 515)	CT-reported acute rib fracture (n = 224)	P value
pH*	7.44 [7.41–7.46]	7.42 [7.39–7.45]	.001
Partial pressure of oxygen (mm Hg)*	75.0 [64.0–86.5]	76.0 [63.0–93.0]	.559
Lactate (mmol/L)†	1.1 [0.8–1.8]	1.5 [0.8–2.6]	.006
White blood cell count (×103/μL)‡	9.3 [7.1–12.4]	9.9 [7.6–13.8]	.044
Hemoglobin (g/dL)‡	12.7 [11.4–14.0]	13.0 [12.0–14.1]	.058
C-reactive protein (mg/dL)§	0.2 [0.1–1.8]	0.2 [0.1–0.8]	.053
Creatinine (mg/dL)‖	0.8 [0.7–1.0]	0.9 [0.7–1.1]	.193
Blood urea nitrogen (mg/dL)‖	16.0 [13.0–21.5]	16.0 [12.0–20.0]	.312
Aspartate aminotransferase (IU/L)‖	29.0 [24.0–38.0]	36.0 [27.0–52.0]	<.001
Alanine aminotransferase (IU/L)‖	19.0 [14.0–27.0]	23.0 [15.5–36.0]	.001
Alkaline phosphatase (IU/L)‖	74.0 [59.0–92.0]	74.0 [61.0–90.5]	.689
Calcium (mg/dL)‖	9.1 [8.8–9.4]	9.0 [8.8–9.3]	.021
Phosphorus (mg/dL)‖	3.0 [2.5–3.4]	3.0 [2.7–3.6]	.111
Sodium (mmol/L)¶	138.0 [136.0–140.0]	138.0 [136.0–140.0]	.587
Chloride (mmol/L)#	104.0 [102.0–107.0]	105.0 [101.0–106.0]	.928
Albumin (g/dL)‖	4.0 [3.8–4.3]	4.0 [3.7–4.2]	.647

Data are presented as median [interquartile range] and use available observations; laboratory testing was clinician-directed and was not obtained for every patient.

Other laboratory variables (‡–#) thus had variable-specific denominators of 564 to 568 patients in total (390–393 without and 174–175 with acute fracture).

All tests used unrounded observations; pH is displayed to 2 decimal places because the between-group difference is smaller than 0.1 pH unit; all other measures are displayed to one decimal place.

CT = computed tomography, IU = international unit.

^*pH and partial pressure of oxygen, 263/99.

^†Lactate, 246/94.

^‡White blood cell count and hemoglobin, 393/175.

^§C-reactive protein, 390/174.

^‖Creatinine, blood urea nitrogen, aspartate aminotransferase, alanine aminotransferase, alkaline phosphatase, calcium, phosphorus, and albumin, 392/175.

^¶Sodium, 391/174.

^#Chloride, 390/175.

Table 3.

Univariable and multivariable logistic regression analyses of clinical factors associated with acute rib fracture on chest computed tomography.

Variable	Univariable analysis OR (95% CI)	P value	Multivariable analysis aOR (95% CI)	P value
Age 65–74 yr	Reference	–	Reference	–
Age 75–84 yr	1.530 (1.088–2.152)	.015	2.236 (1.443–3.465)	<.001
Age ≥85 yr	3.338 (1.978–5.635)	<.001	8.391 (4.354–16.172)	<.001
Male sex (vs female)	2.191 (1.592–3.014)	<.001	1.791 (1.187–2.701)	.005
Fall from an elevated surface (vs ground-level slip/trip)	2.516 (1.823–3.474)	<.001	2.415 (1.561–3.734)	<.001
Non-alert mental status (vs alert)	2.016 (1.219–3.334)	.006	1.891 (1.017–3.514)	.044
Head trauma (yes vs no)	1.908 (1.386–2.625)	<.001	1.418 (0.929–2.165)	.105
Localized chest wall pain (yes vs no)	10.111 (6.932–14.748)	<.001	11.541 (7.584–17.562)	<.001

The standard maximum-likelihood primary model included all 739 patients and 224 outcome events. Six clinical variables, represented by 7 regression coefficients, were selected on clinical grounds independently of univariable P values, entered simultaneously, and retained; no automated variable selection was used. The model was defined for this analysis and was not prospectively specified before data collection. aORs are mutually adjusted for all displayed variables. The numerical fall-height field, selectively ordered laboratory variables, and rare sternal and clavicular pain findings were not included in the primary model. No patient was excluded at model entry, and no missing value was imputed. Sensitivity analyses are reported in Table S3, Supplemental Digital Content 4. P values are based on full-precision coefficients; displayed estimates are rounded. Group-comparison P values in Table 1 may differ from the logistic-regression P values shown here because that analysis applies a continuity correction to the 2 × 2 comparisons.

aOR = adjusted odds ratio, CI = confidence interval, OR = odds ratio.

The primary multivariable logistic regression used standard maximum likelihood and included 6 clinical variables represented by 7 regression coefficients: age category (65–74 years as the reference, 75–84 years, and ≥85 years), male sex, fall from an elevated surface, non-alert mental status, head trauma, and localized chest wall pain (Table S1, Supplemental Digital Content 3). The variables were selected on clinical grounds independently of univariable P values, entered simultaneously, and retained; no backward, forward, or stepwise selection was used. KTAS level, oxygen therapy, dyspnea, and vital signs were not included as covariates because they are physiologic or triage responses that may be consequences of the injury itself or closely tied to the clinician’s decision to obtain CT. Adjustment for such variables could introduce overadjustment or collider bias. Head trauma was retained as a marker of impact distribution documented at presentation rather than as a physiologic response. Current smoking, other location-specific pain findings, comorbidities, and preinjury medications were likewise not entered. The primary analysis was deliberately restricted to a small set of core presentation-level factors to preserve a parsimonious model with an adequate number of outcome events per coefficient; the exclusion of these variables reflected this rationale rather than their univariable P values. Under the symptom-field coding convention described above, the 6 covariates had an analytic value for all 739 patients, including 224 patients with the outcome, and no patient was excluded at model entry. The primary covariate set was not specified before data collection.

Laboratory variables were not included in the primary model because testing was clinician-directed, and missing values were not imputed. For subgroup analyses, ABG testing was defined by the availability of arterial pH and partial pressure of oxygen results; lactate was unavailable in 22 of these 362 patients. Patients with and without ABG testing were compared descriptively using standardized mean differences and P values (Table S2, Supplemental Digital Content 2). Sensitivity analyses modeled age continuously and with a restricted cubic spline using 3 knots at the 10th, 50th, and 90th percentiles; applied Firth penalization to the primary model; added sternal and clavicular pain together in an extended Firth model; refitted the clinical model separately in the ABG-tested and untested subgroups; and added pH and lactate among patients with both measures (Table S3, Supplemental Digital Content 4). Results are presented as odds ratios (ORs) for univariable analyses and adjusted odds ratios (aORs) for multivariable analyses, each with 95% CIs; two-sided P < .05 indicated statistical significance.

3. Results

3.1. Patient baseline characteristics

Of 739 patients, 224 had a CT-reported acute rib fracture, and 515 had no CT-reported acute rib fracture (Table 1). The acute-fracture group was older and included more patients aged ≥75 years (55.4% vs 40.6%) and more men (54.9% vs 35.7%). Falls from an elevated surface were more common (52.2% vs 30.3%). The acute-fracture group also had more urgent (numerically lower) KTAS levels,^[33] more non-alert mental status (13.8% vs 7.4%), and more oxygen use (42.0% vs 21.4%). Current smoking was more common (17.9% vs 10.7%; P = .010), whereas recorded comorbidities and most medication variables did not differ substantially.

Head trauma was more common in the acute-fracture group (49.1% vs 33.6%). Localized chest wall pain showed the largest difference (57.1% vs 11.7%; P < .001). Sternal pain (4.9% vs 1.6%; P = .016), dyspnea (19.2% vs 12.8%; P = .033), clavicular pain (12.9% vs 2.3%; P < .001), and shoulder pain (21.4% vs 14.0%; P = .016) were also more common, whereas lower back pain was less common (11.6% vs 21.0%; P = .003). Other location-specific pain findings did not differ significantly (Table 1).

3.2. Arterial blood gas and laboratory findings

ABG pH and partial pressure of oxygen were available for 263 patients with no CT-reported acute rib fracture and 99 with acute rib fracture; lactate was available for 246 and 94 patients, respectively. Other laboratory measures were available for 564 to 568 patients (390–393 with no acute fracture and 174–175 with acute fracture; Table 2). In available-case analyses, pH was lower and lactate higher in the acute-fracture group. White blood cell count and aspartate and alanine aminotransferase levels were higher, whereas calcium was slightly lower; the remaining measures did not differ significantly (Table 2). Compared with the 377 patients without ABG testing, the 362 tested patients more often had non-alert mental status and received oxygen therapy, and less often had localized chest wall pain. The fracture proportion was 27.3% in tested patients and 33.2% in untested patients (P = .102; Table S2, Supplemental Digital Content 2).

3.3. Factors associated with occult acute rib fracture

The primary multivariable model included all 739 patients and 224 outcome events. Relative to ages 65 to 74 years, the aOR was 2.236 (95% CI, 1.443–3.465; P < .001) for ages 75 to 84 years and 8.391 (95% CI, 4.354–16.172; P < .001) for ages ≥85 years. Male sex (aOR, 1.791; 95% CI, 1.187–2.701; P = .005), fall from an elevated surface (aOR, 2.415; 95% CI, 1.561–3.734; P < .001), non-alert mental status (aOR, 1.891; 95% CI, 1.017–3.514; P = .044), and localized chest wall pain (aOR, 11.541; 95% CI, 7.584–17.562; P < .001) were associated with CT-reported acute rib fracture after adjustment. Head trauma showed no statistically significant association (aOR, 1.418; 95% CI, 0.929–2.165; P = .105; Table 3).

In sensitivity analyses, continuous age yielded an aOR of 1.106 per year (95% CI, 1.074–1.138; P < .001), equivalent to 2.728 per 10 years (95% CI, 2.036–3.656), and the 3-knot restricted cubic spline model showed no evidence of a nonlinear component relative to a linear age effect (likelihood-ratio χ² = 0.02, 1 df; P = .875). Firth estimates for the primary model were similar. In the extended Firth model, the adjusted estimate for sternal pain was imprecise and not statistically significant (aOR, 1.698; 95% profile-likelihood CI, 0.524–5.319; P = .370), whereas clavicular pain showed a statistically significant but imprecise adjusted association (aOR, 7.657; 95% profile-likelihood CI, 3.415–17.773; P < .001). In the continuous-age model, the estimate for non-alert mental status was attenuated, and its CI included the null (aOR, 1.784; 95% CI, 0.958–3.324; P = .068). Clinical estimates were broadly similar in the ABG-tested and untested subgroup refits and after adding pH and lactate; neither laboratory variable was statistically significant in this selectively tested subgroup model (Table S3, Supplemental Digital Content 4).

4. Discussion

Among 739 older fall patients selected for CT after negative initial CXR, 224 (30.3%) had a CT-reported acute rib fracture. This is a detection proportion within a clinician-selected cohort, not the prevalence among all older fall patients with negative CXR or the false-negative rate of CXR. Older age, male sex, fall from an elevated surface, non-alert mental status, and localized chest wall pain were associated with CT-reported acute rib fracture after adjustment; the association with non-alert mental status was less stable across age specifications.

The primary analysis used a compact clinical model, entered all covariates together without automated selection, and retained all 739 patients. Physiologic and triage variables such as oxygen therapy, dyspnea, and triage level were intentionally omitted from the primary model because they may represent consequences of thoracic injury or determinants of CT selection rather than baseline characteristics. Adjusting for them could have attenuated true associations or introduced collider bias. Head trauma, in contrast, was retained as a presentation-level marker of impact distribution. Categorizing age in the primary model avoided imposing a constant per-year effect, while continuous and spline analyses supported the direction and increasing magnitude of the age association. The adjusted age estimates exceeded the unadjusted estimates because older patients were less often male (49.5%, 34.1%, and 23.2% across the 3 age categories) and less often fell from an elevated surface (46.3%, 25.8%, and 24.6%), whereas both factors were positively associated with fracture. The crude age association was therefore negatively confounded, and adjustment unmasked a steeper age gradient rather than creating one. The Firth analyses produced similar central estimates and allowed the 2 uncommon focal findings to be evaluated without adding them to the primary model. These analyses address model stability but do not convert the study into a prediction analysis or establish causality.

Previous research in older fall patients has emphasized the yield of CT-only fractures and their associations with resource use or in-hospital outcomes,^[9] whereas studies in broader blunt-trauma populations have examined thoracic injuries and imaging strategies.^[10–12] The American College of Radiology criteria and NEXUS Chest CT address imaging selection more generally,^[5,34] and management guidance concerns patients with established multiple rib fractures.^[7] This study provides a presentation-level description within the narrower pathway of older patients already selected for same-visit CT after CXR without a definite, possible, or suspected acute rib fracture. It did not compare imaging strategies or determine whether CT findings changed treatment or outcomes.

Localized chest wall pain showed the strongest adjusted association, reinforcing the importance of careful palpation when concern persists after CXR.^[4,9,10] This finding cannot be interpreted as a stand-alone indication for CT because every participant had already been selected for CT, and documentation of focal symptoms may itself have influenced that decision. Clavicular pain showed a statistically significant but imprecise association in the penalized sensitivity model based on only 41 patients; the penalized estimate for sternal pain was imprecise and not statistically significant. The documented categorical mechanism was retained because it was clinically interpretable from the event description. The age, sex, mechanism, and mental-status associations may still reflect unmeasured exposure, frailty, injury severity, or selection rather than direct biological effects.

Patients with and without ABG testing differed in several bedside characteristics, consistent with selective, clinician-directed testing. Laboratory values were therefore excluded from the primary model and not imputed. Clinical estimates were broadly similar in direction across the full-cohort model, the ABG-tested and untested subgroup refits, and the pH-and-lactate-augmented model. These exploratory comparisons cannot eliminate selection bias or support interpreting laboratory findings as cohort-wide fracture markers.

Strengths include the explicit screening denominator and sequential cohort flow, exclusion of equivocal CXR reports, retention of one index visit per patient, detailed description of the clinician-directed CT pathway, and use of all 739 patients in the primary model. Important limitations remain. Clinician-selected CT use likely enriched the cohort for focal symptoms or perceived injury severity, and the 232 encounters without CT did not undergo detailed abstraction or CT outcome verification, precluding a CT-versus-no-CT comparison. The direction and magnitude of selection bias in the individual associations remain uncertain.^[35] The single-center retrospective design limits transportability. The outcome was based on routine, nonblinded final CT reports rather than blinded central review; investigators did not reinterpret images, and interobserver agreement was not assessed. Undocumented or unassessable symptoms were generally coded absent, and independent duplicate abstraction was not performed, so symptom under-ascertainment or misclassification is possible. The primary covariate set was selected for this analysis and was not specified before data collection. Although no P value threshold or automated procedure was used to choose the covariates, the analysis was necessarily conducted with knowledge of the univariable results, so covariate selection cannot be regarded as fully independent of the observed data. Issues of multiple comparisons, residual confounding, and imprecision for uncommon findings remain. Clinical management and patient-centered outcomes were not assessed.

Future studies should use consecutive enrollment or a prespecified verification strategy, prospectively recorded CT indications, and standardized symptom and mechanism definitions, and should assess patient-centered outcomes. Explanatory or prediction analyses should prespecify a parsimonious covariate set, address missingness and functional form, use sparse-data methods when needed, and undergo appropriate validation before informing CT selection.

5. Conclusion

Among older fall patients selected for same-visit chest CT after negative initial CXR, older age, male sex, fall from an elevated surface, and localized chest wall pain were associated with CT-reported acute rib fracture after adjustment in the primary full-cohort model. The association with non-alert mental status was less stable across age specifications. These findings characterize associations within a clinician-selected imaging pathway and do not define a CT decision rule, demonstrate clinical benefit from CT, or support routine CT after negative CXR.

Author contributions

Conceptualization: Won Young Sung.

Data curation: Joo Han Kwon.

Formal analysis: Won Young Sung.

Investigation: Joo Han Kwon.

Methodology: Joo Han Kwon, Won Young Sung.

Project administration: Joo Han Kwon.

Software: Won Young Sung.

Supervision: Won Young Sung.

Validation: Won Young Sung.

Writing – original draft: Joo Han Kwon, Won Young Sung.

Writing – review & editing: Joo Han Kwon, Won Young Sung.