Point-of-care ultrasound for emergency pericardiocentesis: a multidimensional CUSUM analysis of the learning curve

Abstract

Background

Emergency point-of-care ultrasound-guided percutaneous pericardial catheter drainage (US-GPPCD) is a life-saving intervention for cardiac tamponade, yet the learning curve among novice operators in emergency settings remains poorly characterized, especially for multidimensional metrics and operator positioning. This study aimed to evaluate the learning curve, procedural efficiency, operator confidence, complications, and the influence of operator positioning for US-GPPCD performed by a single novice operator under direct supervision.

Methods

A total of 33 consecutive patients undergoing US-GPPCD between October 2024 and December 2025 by a single novice operator without prior independent experience were enrolled. Procedural metrics were prospectively recorded immediately post-procedure, while clinical data were retrospectively collected. Demographic and clinical characteristics, procedure time, needle insertion site localization time, operator confidence score, operator position, and complications were analyzed. Three separate cumulative sum (CUSUM) learning curves were constructed for multidimensional learning curve analysis.

Results

CUSUM analysis divided the learning curve into two phases: phase 1 (cases 1–14, skill-acquisition) and phase 2 (cases 15–33, skill-proficiency). Procedure time and localization time were significantly longer, and operator confidence score was significantly lower, in phase 1 than in phase 2 (all P < 0.01). The complication rate was higher in phase 1 (28.6%) than in phase 2 (5.3%), with no statistically significant difference (P = 0.14). While left-side positioning was associated with shorter procedure time in unadjusted analysis, this effect was confounded by operator experience and not significant after adjustment.

Conclusion

In this exploratory analysis, US-GPPCD showed acceptable short-term safety under direct supervision. A significant inflection point in procedural efficiency was observed at approximately 14 cases, while operator confidence consolidated shortly thereafter. These findings provide preliminary, hypothesis-generating evidence that may inform future training protocols, but require validation across multiple operators and centers.

Introduction

Pericardial effusion is a common clinical emergency, and large or rapidly accumulating pericardial effusions can lead to cardiac tamponade, causing a sudden drop in cardiac output and posing a serious threat to patient life [1]. Emergency point-of-care ultrasound-guided percutaneous pericardial catheter drainage (US-GPPCD) is the first-line lifesaving procedure for relieving acute cardiac tamponade [2, 3]. In contrast to traditional blind or electrocardiogram-guided approaches, US-GPPCD provides real-time visualization, enabling precise avoidance of vital structures (e.g., heart, coronary arteries) and significantly improving procedural safety and accuracy. It has become the preferred intervention in emergency, cardiovascular, and critical care medicine [3–5]. However, this technique must be performed in a dynamically beating cardiac environment within a confined anatomical space adjacent to vital organs, requiring operators to possess integrated skills in real-time ultrasound localization, fine needle manipulation, and emergency management [6, 7].

Similar to other ultrasound-guided percutaneous techniques [8, 9], mastering US-GPPCD involves a substantial learning curve for novice residents. This process demands not only technical proficiency in needle manipulation but also accurate real-time ultrasound interpretation, requiring extensive clinical practice and structured feedback to achieve procedural competency. Cumulative sum (CUSUM) analysis serves as a reliable tool in medical education for evaluating the learning curve of procedural skills [8, 9]. By plotting the cumulative deviation of performance from a predefined target level, it graphically depicts the trajectory of skill progression and provides an objective basis for quantifying proficiency.

Despite the utility of CUSUM analysis, significant gaps remain in the existing research regarding the learning curve for this technique. Previous studies have primarily focused on its safety and efficacy [9], with few systematic analyses of the learning curve specifically in the context of emergency point-of-care settings and single-operator consecutive procedures. Most existing studies are exploratory with small sample sizes, lacking a comprehensive quantitative assessment of factors such as procedure time, complication rate, and operator confidence. Furthermore, the minimum number of cases required for a novice to achieve procedural competency has not been clearly defined. Additionally, operator positioning—a potential key factor influencing procedural efficiency—has not been incorporated into correlative analyses of the learning curve, leaving a critical gap in robust evidence to support standardized, stratified training protocols for emergency US-GPPCD.

Based on the aforementioned clinical needs and research gaps, this study aimed to employ CUSUM analysis to systematically quantify the learning process of US-GPPCD by a single novice resident. Using procedure time as the primary metric and incorporating complication rates and operator confidence score, this study aimed to multi-dimensionally define the case threshold for achieving procedural competency. Furthermore, for the first time, this study investigated the impact of different operator positioning strategies on procedural outcomes across different phases of the learning curve. These findings provide a quantitative basis for developing stratified and precise US-GPPCD training protocols, as well as practical support for the standardization and optimization of clinical procedural workflows.

Methods

Study design

This was a retrospective analysis of a prospectively maintained procedural log, consecutively including 33 patients who underwent ultrasound-guided percutaneous pericardial catheter drainage (US-GPPCD) at Zhangzhou Affiliated Hospital of Fujian Medical University between October 2024 and December 2025. All procedures were performed consecutively by a single novice ultrasound resident with 2 years of point-of-care ultrasound experience and 1 year of observational training in pericardiocentesis but no prior independent US-GPPCD experience, under the direct real-time supervision of a senior interventional ultrasound physician with 10 years of experience and a pre-specified emergency takeover protocol to ensure patient safety. The study was approved by the Ethics Review Committee of Zhangzhou Affiliated Hospital of Fujian Medical University (Approval No.: 2026LWB047) and conducted in strict accordance with the Declaration of Helsinki, with a waiver of informed consent granted due to the observational design, minimal patient risk, and anonymized data collection. Inclusion criteria were: requirement for emergency US-GPPCD for cardiac tamponade or massive pericardial effusion; procedure performed by the same novice operator; and complete procedural and 7-day follow-up data available. Exclusion criteria included: infection or skin breakdown at the puncture site; severe coagulation dysfunction (International Normalized Ratio (INR) > 1.5, platelet count < 50 × 10⁹/L); pericardial effusion volume < 50 mL; and patient or family refusal of the procedure.

Procedural metrics including procedure time, needle insertion site localization time, operator position, and operator confidence score were recorded prospectively via a dedicated anonymized log immediately post-procedure, with procedure time defined from verbal procedure start to wound closure, localization time from probe placement to optimal puncture site confirmation, and operator confidence score assessed on a 1–10 scale immediately after each procedure. Patient demographic and clinical characteristics, procedural indications, puncture site, complications, and 7-day postoperative follow-up data were retrospectively collected from electronic medical records, with all patients completing standardized short-term follow-up for complications and clinical outcomes.

Operator background

The novice operator was a third-year resident physician in the Department of Ultrasound, with no prior independent experience in US-GPPCD. He had completed standardized ultrasound training and undergone observational and simulation training for pericardial access. This study represented his first independent performance of the procedure.

Supervision and safety protocol

All procedures were performed under the direct supervision of an attending physician with > 10 years of experience in interventional ultrasound. The supervisor was on-site throughout the procedure and was prepared to take over management at any time. Predefined takeover criteria included: failure to identify an appropriate puncture site within 5 min, two unsuccessful puncture attempts, or any evidence of hemodynamic instability. No intervention takeover was required during the study.

Operator confidence assessment

Operator confidence was assessed using an 11-point numeric rating scale (0–10) immediately after each procedure, with 0 indicating no confidence in independent operation and 10 indicating full confidence in independent operation. This score was prospectively recorded in a dedicated procedural log immediately post-procedure. These data were subsequently retrieved retrospectively together with clinical data from medical records. The score was included as a dimension in the multimetric CUSUM analysis to determine the inflection point of operator confidence along the learning curve.

Procedure for point-of-care US-GPPCD

All procedures were performed using a point-of-care color Doppler ultrasound system (Mindray, Shenzhen, China) equipped with linear (4–12 MHz) and convex (1–5 MHz) array probes. A disposable sterile indwelling drainage catheter set (Model R, 7-Fr (French), 20 cm, 3-lumen; Lingjie Medical, Guangzhou, China) was used for all cases. Under local anesthesia, the pericardial drainage catheter was placed into the pericardial cavity via the Seldinger technique, with continuous real-time point-of-care ultrasound guidance throughout. The key procedural steps were as follows:

1. Pre-procedural Assessment and Positioning: The patient was placed in a semi-recumbent or supine position (head of bed elevated 30°–45°), with the intended puncture site (anterior chest wall or subxiphoid region) exposed. A convex probe (2–5 MHz) was used for a comprehensive scan to identify the deepest pocket of effusion, measure its depth, characterize fluid properties (free-flowing vs. loculated), and mark adjacent vital structures. The shortest and safest needle trajectory was planned to avoid major vessels and areas of excessive cardiac motion.

2. Aseptic Preparation and Anesthesia: The puncture site was disinfected three times with iodophor (area ≥ 15 cm in diameter) and draped sterilely. Local infiltration anesthesia was administered with 2% lidocaine, layered from the skin through the chest wall soft tissue down to the outer surface of the parietal pericardium. The needle tip was visualized in real time under ultrasound during infiltration to prevent intravascular or intrapericardial injection (Fig. 1A).

3. Ultrasound-Guided Puncture: With the ultrasound probe stabilized adjacent to the planned path, an 18 gauge puncture needle was advanced under real time in plane ultrasound guidance. Advancement was slowed as the needle tip approached the parietal pericardium. Successful entry was confirmed by direct ultrasound visualization of the needle tip entering the effusion cavity (Fig. 1B) and aspiration of serous, sanguineous, or purulent fluid, after which the needle was secured.

4. Guidewire Insertion: Under continuous ultrasound monitoring, a J tip guidewire was advanced through the needle into the effusion cavity, with its position confirmed to avoid penetration of the posterior pericardium or myocardial injury (Fig. 1C). After inserting 15–20 cm of the guidewire, the needle was withdrawn while applying local pressure with sterile gauze.

5. Tract Dilation: The subcutaneous tissue was gently dilated over the guidewire up to the outer layer of the parietal pericardium using a vascular dilator, with care taken to avoid excessive depth. The guidewire was secured throughout the procedureto prevent displacement.

6. Catheter Placement and Securing: The drainage catheter was advanced over the guidewire into the pericardial space. Free fluid flow was confirmed upon opening the catheter port. The catheter was secured to the skin with non absorbable sutures, covered with a sterile dressing, and connected to a drainage bag. The insertion date and catheter depth were documented.

7. Post procedural Verification: Immediate point-of-care ultrasound was performed post-procedure to confirm the correct position of the catheter within the pericardial cavity (Fig. 1D) and to rule out complications such as hemopericardium or pneumothorax. The patient's vital signs and drainage output were continuously monitored.

Fig. 1.

Sonographic images of key procedural steps during US-GPPCD. Red arrows indicate critical structures in each panel. (A) Anesthesia phase: The hyperechoic anesthetic needle tip is advanced into the chest wall soft tissue down to the outer surface of the parietal pericardium. (B) Puncture phase: The 18-gauge puncture needle traverses the parietal pericardium into the anechoic pericardial effusion. (C) Guidewire insertion: The J-tip guidewire is positioned within the effusion cavity. (D) Catheter placement: The echogenic drainage catheter is confirmed to be correctly positioned within the effusion cavity

Statistical analyses

CUSUM analysis was used to quantitatively evaluate the learning curve, with CUSUM curves fitted using a cubic polynomial model. Operator position was coded as a binary variable (0 = right side, 1 = left side) for multivariable linear regression analysis. Correlation analyses were performed using Pearson correlation and Spearman’s rank correlation. Comparative analyses were performed using the t-test, Chi-square test, Fisher’s exact test, and Mann-Whitney U-test. The Chow test was applied to determine the inflection point of the learning curve. Statistical significance was defined as P < 0.05. All statistical analyses were performed using IBM SPSS Statistics version 27.0 (IBM Corp., Armonk, NY, USA).

CUSUM analysis

This study employed CUSUM analysis to quantify the learning curve for US-GPPCD across three dimensions: procedural efficiency, localization accuracy, and subjective confidence. Specifically, four CUSUM series were constructed:

1. Procedure Time CUSUM (CUSUM_PT): representing the learning curve of procedural efficiency.

2. Puncture Site Localization Time CUSUM (CUSUM_LT): representing the learning curve of puncture site localization accuracy and speed.

3. Operator Confidence Score CUSUM (CUSUM_CS): representing the learning curve for assessed operator confidence.

4. Complication CUSUM (CUSUM_C): reflecting the change in complication risk across the learning curve (complication was coded as 0 = no complication, 1 = any complication).

The same unified calculation formula was applied to all four CUSUM series:

We acknowledge that this target (µ) represents the operator’s own overall mean during the learning process rather than an expert-derived benchmark; which represents a conservative and practice-based analytical strategy, where n is the sequential case number, x_i is the observed value for the i-th case (procedure time, localization time, operator confidence score, or complication occurrence, respectively), and µ is the mean of the corresponding metric across all included cases. Four learning curves were plotted with the case sequence number on the horizontal axis and the cumulative CUSUM values on the vertical axis. The inflection points of the curves—representing the skill proficiency threshold—were determined in two steps. First, an inflection candidate was initially identified by visual inspection of the slope change. This candidate point was then statistically validated using the Chow test, which assessed whether the regression models of the four metrics differed significantly before and after the inflection point (P < 0.05). The final case threshold required to achieve procedural proficiency was thereby established.

Learning curve phase division

Based on the inflection points identified by CUSUM analysis and validated with the Chow test, the learning curve was divided into two phases: the Skill Acquisition Phase (Phase 1), characterized by highly variable procedural and localization times, low operator confidence, and higher complication risk and frequent deviations above the proficiency threshold; and the Skill Proficiency Phase (Phase 2), characterized by a favorable trends in CUSUM values, indicating improved operational proficiency with minimal deviation from the target.

Results

Baseline clinical characteristics by learning curve phase

A total of 33 consecutive patients undergoing US-GPPCD were enrolled and divided into the skill acquisition phase (Phase 1, n = 14) and skill proficiency phase (Phase 2, n = 19) according to the learning curve inflection point confirmed by the Chow test. Baseline characteristics are presented as median (interquartile range) or number (%) (Table 1). Between-group comparisons were performed using the Mann-Whitney U test and chi-square test. No significant differences were observed in age, height, weight, BMI, sex, indication, puncture approach, or severity of pericardial effusion (all P > 0.05). Operator position differed significantly between phases (P = 0.02), with most procedures in Phase 1 performed from the patient’s right side (71.4%) and those in Phase 2 from the left side (68.4%). Baseline profiles were generally well balanced, and operator position was adjusted for in subsequent multivariable analyses to control for confounding.

Table 1.

Baseline characteristics and procedural outcomes according to learning curve phases

Variables	Overall (n = 33)	Phase 1 (n = 14)	Phase 2 (n = 19)	Statistic	P-value
Continuous variables, median (IQR)
Age, years	66.0 (58.5–72.0)	64.5 (57.0–70.5)	67.0 (59.0–72.0)	U = 155.50	0.42
Height, cm	163.0 (155.5–170.0)	162.5 (158.0–168.0)	167.0 (155.0–170.0)	U = 139.50	0.82
Weight, kg	57.0 (50.0–61.5)	56.0 (53.0–59.0)	57.0 (46.0–62.0)	U = 137.00	0.90
BMI, kg/m²	21.3 (18.8–23.4)	21.3 (20.7–23.4)	21.0 (16.5–23.4)	U = 115.50	0.53
Categorical variables, n (%)
Sex				χ2 = 1.34	0.25
Male	18 (54.5%)	6 (42.9%)	12 (63.2%)
Female	15 (45.5%)	8 (57.1%)	7 (36.8%)
Indication				χ2 = 3.48	0.06
Oncologya	18 (54.5%)	5 (35.7%)	13 (68.4%)
Othersb	15 (45.5%)	9 (64.3%)	6 (31.6%)
Puncture approach				χ2 = 0.04	0.98
Parasternal	23 (69.7%)	10 (71.4%)	13 (68.4%)
Subxiphoid	5 (15.2%)	2 (14.3%)	3 (15.8%)
Apical	5 (15.2%)	2 (14.3%)	3 (15.8%)
Operator position				χ2 = 5.13	0.02
Right side	17 (51.5%)	4 (28.6%)	13 (68.4%)
Left side	16 (48.5%)	10 (71.4%)	6 (31.8%)
Severity of Pericardial Effusion				χ2 = 2.89	0.24
Mild	4 (12.1%)	3 (21.4%)	1 (5.3%)
Moderate	19 (57.6%)	6 (42.9%)	13 (68.4%)
Severe	10 (30.3%)	5 (35.7%)	5 (26.3%)
Complications
Overall Complication Rate	5 (15.2%)	4 (28.6%)	1 (5.3%)	χ2 = 3.41	0.14
Puncture site hematoma	2 (6.1%)	2 (14.3%)	1 (5.3%)
Catheter displacement	1 (3.0%)	1 (7.1%)	0 (0%)
Catheter occlusion	2 (6.1%)	1 (7.1%)	0 (0%)

Data are presented as median (interquartile range, IQR) or number (%). Continuous variables were compared using the Mann-Whitney U test (U values reported); categorical variables were compared using the chi-square test (χ²values reported). Phase 1: skill acquisition phase; Phase 2: skill proficiency phase; BMI: body mass index

ᵃOncology-related diseases (18 cases, 54.5%): Malignant lung tumor (12 cases, 36.4%), malignant breast tumor (2 cases, 6.1%), esophageal cancer (1 case, 3.0%), gastric cancer (1 case, 3.0%), secondary malignant tumor (1 case, 3.0%), and other malignant tumor (1 case, 3.0%)

ᵇOther diseases (15 cases, 45.5%): Acute heart failure (2 cases, 6.1%), trauma (2 cases, 6.1%), and 11 other conditions (e.g., chronic disease, respiratory failure, pulmonary tuberculosis, mediastinal tumor, etc.) (1 case each, 3.0% each)

Trend and influencing factors of procedure time

Among the 33 included patients, the median procedure time was 8.7 min. A declining trend in procedure time was observed with increasing case sequence number (Fig. 2). Pearson correlation analysis identified a strong negative linear correlation between procedure time and case sequence number (r = − 0.92, P < 0.01), with a linear regression goodness-of-fit of R²= 0.845. No significant linear correlations were found between procedure time and patient age (r = − 0.04, P = 0.81), body weight (r = 0.05, P = 0.78), height (r = − 0.05, P = 0.77), or BMI (r = 0.13, P = 0.47).

Fig. 2.

Scatter plot of case number vs. procedure time with a fitted linear regression line. A clear negative linear association was observed between cumulative experience and procedure time (Pearson r = − 0.92, P < 0.01)

CUSUM analysis results

CUSUM analysis was performed separately for procedure time, needle insertion site localization time, and operator confidence score. Each metric was fitted with a cubic polynomial model. The goodness of fit was high for all three indicators (R² > 0.98), and the overall regression effects were statistically significant (P < 0.001) (Table 2). The target value (µ) in the CUSUM formula was defined as the mean value across all 33 cases, not an expert-derived proficiency benchmark.

Table 2.

CUSUM analysis model parameters and inflection points

Metric	Fitted Equation	R²	P-value	Inflection Point (Case Number)
CUSUMPT		0.995	< 0.001	14
CUSUMLT		0.993	< 0.001	14
CUSUMCS		0.987	< 0.001	16

R²: coefficient of determination; P-value from overall regression test of cubic polynomialmodel

The fitted curve for CUSUM_PT (Fig. 3A) peaked at the 14th case. Given the modest sample size, this threshold should be interpreted as exploratory. This case number was used to divide the learning curve into Phase 1 (cases 1–14) and Phase 2 (cases 15–33). The fitted curve for CUSUM_LT (Fig. 3B) also peaked at the 14th case, indicating an inflection point identical to that of the procedure time metric. The fitted curve for CUSUM_CS (Fig. 3C) peaked at the 16th case, with its inflection point lagging slightly behind that of the procedure time metrics.

Fig. 3.

CUSUM learning curves for US-GPPCD: procedure time (CUSUM_PT, A), localization time (CUSUM_LT, B), and operator confidence score (CUSUM_CS, C). The inflection points at case 14 (A, B) and case 16 (C) separated the skill-acquisition phase (Phase 1) from the skill-proficiency phase (Phase 2)

Case interval analysis

The median interval between consecutive procedures was 12 days (IQR 10–15 days). No significant correlation was found between the interval length and the change in procedure time (Spearman’s ρ = −0.12, P = 0.53), likely due to the uniformly low case frequency.

Comparison of core outcome metrics across learning curve phases

To validate the identified learning curve inflection point and quantify changes in procedural skills before and after the breakpoint, this study applied the Chow test to assess structural breaks in the learning curves for core operative and subjective metrics. Statistically significant structural breaks were confirmed for all continuous outcomes (procedure time, needle insertion site localization time, operator confidence score; all P < 0.001) (Table 3). A significant reduction in postoperative complications was also observed in Phase 2 versus Phase 1 (LR χ² = 9.8, P = 0.044). These findings align with the progressive improvement in procedural efficiency and operator confidence visualized in the CUSUM learning curves, confirming the validity of the 14-case inflection point.

Table 3.

Key outcomes before and after the learning inflection point

Measured Outcome	Phase 1 (n = 14)	Phase 2 (n = 19)	Statistic	df	P-Value
	Median (IQR) / n (%)	Median (IQR) / n (%)
Procedure time (min)	12.7 (11.5–14.5)	7.9 (7.5–8.5)	F = 20.8	(4, 25)	< 0.001
Needle insertion site localization time (min)	3.9 (3.5–4.3)	2.0 (1.6–2.2)	F = 18.1	(4, 25)	< 0.001
Operator Confidence Score	5.0 (4.0–5.0)	8.0 (7.0–9.0)	F = 9.1	(4, 25)	< 0.001
Postoperative complications	4 (28.6%)	1 (5.3%)	LR χ² = 9.8	4	0.044

For continuous outcomes, the Chow test based on simple linear regression was used to test the structural break of the learning curve. For the binary outcome (postoperative complications), the Chow-type likelihood ratio test based on simple binary logistic regression was applied. IQR, interquartile range; LR, likelihood ratio

Subgroup analysis by effusion severity

To examine whether the observed learning effect was confounded by case complexity, we stratified the analysis by pericardial effusion severity (mild, moderate, severe). As shown in Table 1, the distribution of effusion severity did not differ significantly between Phase 1 and Phase 2 (P = 0.24). Notably, Phase 2 had a lower proportion of both the most technically demanding mild effusions (5% vs. 21%) and the least challenging severe effusions, with a corresponding higher proportion of moderate effusions (68% vs. 43%).

In the moderate effusion subgroup (n = 19), median procedure time decreased significantly from 12.7 min to 7.9 min (P < 0.001). In the severe effusion subgroup (n = 10), median procedure time also decreased substantially from 14.5 min to 8.2 min (P = 0.008). The mild effusion subgroup (n = 4) was underpowered for formal statistical comparison, but the single Phase 2 case (7.6 min) was markedly lower than the three Phase 1 cases (median 11.5 min) (Table 4).

Table 4.

Procedure time by effusion severity and learning phase

Severity	Phase 1 median (IQR), n	Phase 2 median (IQR), n	P-Value
Mild	11.5 (11.2–12.0), n = 3	7.6, n = 1	—
Moderate	12.7 (11.8–13.8), n = 6	7.9 (7.5–8.4), n = 13	< 0.001
Severe	14.5 (12.4–15.5), n = 5	8.2 (7.3–8.7), n = 5	0.008

P values were calculated using the Mann-Whitney U test; data are median (interquartile range).The symbol “—” indicates no formal statistical test was performed due to the small sample size (n = 1 in Phase 2 for the mild effusion subgroup)

These findings collectively indicate that the observed learning effect was consistent across all effusion severity subgroups, and the improved procedural efficiency was not driven by a preferential shift toward technically simpler cases.

Multivariable linear regression analysis for procedure time

To further assess the independent effect of cumulative experience on procedure time, we performed a multivariable linear regression with procedure time as the dependent variable and case number (continuous), effusion severity (categorical: mild/moderate/severe), and operator position (binary: 0 = right, 1 = left) as independent variables. After adjustment, case number remained a significant independent predictor of procedure time (β = −0.273, 95% CI: −0.328 to − 0.218, P < 0.001). Effusion severity (β = 0.025, 95% CI: −0.747 to 0.797, P = 0.947) and operator position (β = −0.288, 95% CI: −1.226 to 0.650, P = 0.535) were not statistically significant, indicating that the unadjusted association between left-side positioning and shorter procedure time was confounded by operator experience (Table 5).

Table 5.

Multivariable linear regression analysis for procedure time

Variable	β (95% CI)	P-value
Case number (per 1-case increase)	-0.273 (–0.328 to − 0.218)	< 0.001
Effusion severity (mild/moderate/severe)	0.025 (–0.747 to 0.797)	0.947
Operator position (0 = right, 1 = left)	-0.288 (-1.226 to 0.650)	0.535

β  unstandardized regression coefficient, CI confidence interval

Comparison of procedure outcomes by operator position and learning phase

The most frequently used puncture approach for US‑GPPCD in this study was the parasternal approach. The overall distribution of the different puncture approaches did not differ significantly between the two learning phases (P > 0.05). The operator’s procedural position was predominantly on the patient’s left side (17 cases, 51.5%), and the distribution of positions differed significantly between the two phases (P = 0.04). In unadjusted analysis, median procedure time was significantly shorter with the left‑side position than with the right‑side position (P < 0.001) (Table 6). However, after adjustment for case number and effusion severity in multivariable regression, operator position was no longer statistically significant (β = −0.288, 95% CI: −1.226 to 0.650, P = 0.535) (Table 5), indicating confounding by experience. Although the complication rate was lower with the left‑side position (5.9%) than with the right‑side position (25.0%), this difference did not reach statistical significance (P = 0.18) (Table 6).

Table 6.

Comparison of procedure outcomes according to the operator’s position

Variable	Patient’s Right Side (n = 16)	Patient’s Left Side (n = 17)	Statistic	P-value
Procedure time (min)	12.1 (8.6–14.4)	7.9 (7.4–9.7)	U = 39.0	< 0.001
Complications	4 (25.0%)	1 (5.9%)	χ2= 2.34	0.18

Unadjusted comparison (Mann-Whitney U test for procedure time, chi-square test for complications); see multivariable regression for adjusted analysis (P = 0.535)

Perioperative complications, follow‑up, and complication‑related learning curve

All patients received point‑of‑care monitoring for 24–48 h postoperatively and completed a standardized 7‑day follow‑up. A total of 5 peri‑ and early post‑procedural complications occurred, including 3 cases of puncture‑site hematoma, 1 case of catheter obstruction, and 1 case of catheter dislodgement (Table 1). Using the learning‑curve inflection point (case 14) as the cutoff, the complication rate was 28.6% (4/14) in Phase 1 (cases 1–14) and 5.3% (1/19) in Phase 2 (cases 15–33). While the difference between phases did not reach statistical significance (P = 0.14), a clear downward numerical trend was observed.

Complication management differed between the phases: all 4 complications in Phase 1 required active intervention (catheter repositioning or local compression), whereas the single minor hematoma in Phase 2 resolved spontaneously with observation only. The CUSUM curve for complications also peaked at case 14: it rose throughout Phase 1, then declined and stabilized in Phase 2, paralleling the trend of the operative‑skill learning curve (Fig. 4). All US‑GPPCD procedures were successfully completed on the first attempt, with no intra‑procedural failures or cases requiring repeat puncture.

Fig. 4.

CUSUM learning curve for US-GPPCD: complication risk (CUSUM_C). The inflection point at case 14 separated the skill‑acquisition phase (Phase 1) from the skill‑proficiency phase (Phase 2)

Discussion

Conventional approaches to percutaneous pericardial drainage include traditional blind puncture and fluoroscopy-guided techniques [10]; however, US‑GPPCD has emerged as the preferred intervention for emergency cardiac tamponade owing to its real-time visualization capability, and is now widely used in both adult and pediatric critical care settings [11–13]. Studies to date have reported on the safety, efficacy, and preliminary learning curve associated with this technique [14, 15]. However, systematic analyses of the learning curve specifically under point-of-care emergency conditions and involving consecutive procedures performed by a single operator remain scarce. In particular, there is a lack of multidimensional quantitative assessment of operative efficiency, localization accuracy, and operator confidence, which limits the evidence base for developing standardized training protocols for this procedure [16, 17].

This study employed CUSUM analysis to quantitatively assess the learning curve of US-GPPCD across three dimensions: operative efficiency, localization accuracy, and operator confidence. The results showed that the inflection points for both procedure time and puncture site localization time occurred at the 14th case, whereas the inflection point for operator confidence score appeared slightly later, at the 16th case. This reveals a clear learning pattern: stable improvement in operative efficiency precedes the complete establishment of the operator’s subjective confidence [18–20]. This “skill-before-confidence” stepwise pattern aligns closely with that reported by Ma et al. in the learning curve for ultrasound-guided caudal blockade [21]. In terms of the case number required to achieve proficiency, the threshold of 14 cases identified in this study for core operative stability is similar to those reported for other high-risk percutaneous techniques such as ultrasound-guided caudal blockade and pediatric central venous catheterization [21, 22], further supporting the observation that high-risk procedural skills share common learning patterns. All three CUSUM curves were well fitted by cubic polynomials, displaying the characteristic trajectory of a rapid rise to a peak followed by a sustained decline, consistent with the classic pathway of skill acquisition and refinement [23, 24].

Based on these inflection points, the learning curve for US-GPPCD was objectively divided into two phases: a skill-acquisition phase (cases 1–14) and a skill-proficiency phase (cases 15–33). During the skill-acquisition phase, operative and localization times decreased rapidly, marking the swift attainment of core procedural skills. After entering the skill-proficiency phase, the slope of the curves gradually flattened and stabilized, reflecting a transition from basic competency to precise and stable execution, characterized by enhanced emergency decision-making and complication-prevention awareness [3, 7]. This objective, data-driven phase division provides a direct basis for implementing stratified and standardized training: the skill-acquisition phase may emphasize simulation-based exercises to solidify foundational skills, while the skill-proficiency phase should focus on cultivating adaptive competency in complex clinical scenarios.

Subgroup analysis by effusion severity revealed that the reduction in procedure time was consistent across mild, moderate, and severe effusions. Notably, Phase 2 had a higher proportion of moderate effusions (68% vs. 43%) and a lower proportion of mild effusions (5% vs. 21%), indicating that the learning effect was not driven by a shift toward technically simpler cases. This supports the robustness of the observed learning curve.

In unadjusted analysis, left-side operator positioning was associated with shorter procedure time. However, multivariable regression adjusting for case number (experience) and effusion severity showed that this association was not statistically significant after adjustment (β = −0.288, 95% CI: −1.226 to 0.650, P = 0.535), indicating that the observed unadjusted association was largely confounded by operator experience. Nevertheless, from an ergonomic and procedural perspective, the left-side position matches the most commonly used parasternal approach, providing better in-plane needle visualization, smoother probe manipulation, more comfortable wrist posture, and more intuitive hand-eye coordination. These practical advantages may support operator comfort during learning, even though no independent effect on procedure time was detected after adjustment. These potential benefits warrant further investigation in larger, controlled studies. At present, the independent effect of operator position on efficiency could not be established in this study.

Regarding safety, the overall complication rate in this study was 15.2% (5/33). Using the 14th case as the cutoff, the complication rate was 28.6% (4/14) during the skill-acquisition phase and 5.3% (1/19) during the skill-proficiency phase. Although the difference between phases was not statistically significant (P = 0.14), a clear downward numerical trend was observed. The non-significant P-value for complication reduction (P = 0.14) likely reflects insufficient statistical power given the sample size. The change in complication management style (all Phase 1 complications required active intervention, whereas the single Phase 2 complication resolved spontaneously) provides clinically meaningful supplementary evidence. This suggests that accumulated experience may not only reduce the incidence of complications but also optimize the strategy for their management. All procedures were successfully completed on the first attempt, with no intraprocedural failures or cases requiring repeat puncture, reaffirming the favorable safety profile of US-GPPCD when performed under strict procedural standards.

The case interval analysis showed a median of 12 days between procedures and no significant correlation between interval length and improvement in procedure time (Spearman’s ρ = −0.12, P = 0.53). This suggests that despite the low case frequency (approximately 2 cases per month), the observed learning curve was still evident, implying that skill acquisition may not be heavily dependent on training density in this low-volume emergency procedure.

This study has several limitations. First, its single-center, single-operator design and modest sample size (n = 33) limit generalizability; moreover, the CUSUM target (µ) was based on the operator’s own mean performance rather than an expert benchmark, representing a conservative approach. Our findings should therefore be considered hypothesis-generating and require validation across multiple operators and centers. Second, case density was uniformly low (approximately 2 cases per month), which precluded assessment of the effect of training intensity on skill acquisition. Third, the postoperative follow-up period was limited to 7 days, which may miss delayed complications. Fourth, although we adjusted for effusion severity in subgroup analyses, other unmeasured factors (e.g., loculation, patient cooperation) and the self-reported nature of operator confidence may introduce bias. Nevertheless, the balanced distribution of measured severity between phases suggests that case-mix bias is unlikely to explain the observed learning effect. Despite these limitations, this study provides a detailed, multidimensional analysis of the US-GPPCD learning curve in a real-world emergency setting, offering preliminary evidence to inform future training protocols. Future multi-center, large-sample, prospective studies with longer follow-up are needed to validate our conclusions.

Conclusions

This CUSUM‑based analysis of a single novice operator suggests that the learning curve for US‑GPPCD has an inflection point at approximately 14 cases, delineating a skill‑acquisition phase from a skill‑proficiency phase. Core operative efficiency improved significantly after crossing this threshold. US‑GPPCD demonstrated an acceptable safety profile under direct supervision. These findings provide preliminary, hypothesis‑generating evidence that may inform the design of phased training programs, but they require validation across multiple operators and centers before being adopted as a quantitative framework.

Abbreviations

US-GPPCD

Emergency point-of-care ultrasound-guided percutaneous pericardial catheter drainage

CUSUM

Cumulative Sum

INR

International Normalized Ratio

CUSUM_PT

Procedure time cumulative sum

CUSUM_LT

Puncture site localization time cumulative sum

CUSUM_CS

Operator confidence score cumulative sum

CUSUM_C

Complication cumulative sum

Authors’ contributions

All authors contributed to the study conception and design. M.C.conceived the research question and administered all processes for this study, X.T.analyzed the data and wrote the manuscript, X.Y.devised the technical methods for the study and critically reviewed the manuscript, Y.C.formally explained the study and obtained informed consent from the participants, C.Z.reviewed the draft manuscript. all authors reviewed the manuscript.

Funding

This study was supported by the Fujian Medical University Startup Fund Grant (2024QH134) from the Fujian Medical University, and the National Key R&D Program of China (2025YFC2423004) funded by the Ministry of Science and Technology of the People’s Republic of China. The funders had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Data availability

The datasets generated and/or analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This study was conducted in accordance with the Declaration of Helsinki. Ethical approval was obtained from the Ethics Committee of the Zhangzhou Hospital Affiliated to Fujian Medical University (Approval No.2026LWB047). Written informed consent was obtained from all participants prior to their inclusion in the study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.