Risk prediction models for complications after flap repair surgery: a systematic review and meta-analysis

Abstract

Objective

To systematically evaluate the performance and applicability of risk prediction models for complications after flap repair and to provide guidance for building and refining models.

Methods

PubMed, Embase, Web of Science, the Cochrane Library, CNKI, SinoMed, VIP and Wanfang were searched for studies on risk prediction models for flap complications. The search period is from inception to December 28, 2024. The PROBAST tool was used to evaluate the quality of the prediction model research, and Stata 18 software was employed to meta-analyze the predictors of the models.

Results

A total of 16 studies were included, 28 risk prediction models were constructed, and the area under the receiver operating characteristic curve (AUC) ranged from 0.655 to 0.964, with 16 prediction models performing well (AUC > 0.7). Eleven articles underwent model calibration, 16 were validated internally, and 3 were validated externally. The results of the PROBAST review revealed that all 16 studies were at high risk of bias. The incidence rate of flap complications was 14.8% (95% CI, 10.7 − 19.0%). Body mass index (BMI), smoking history, long flap reconstruction time, diabetes mellitus, hypertension, and postoperative infection were independent risk factors for complications after flap repair (P < 0.05).

Conclusion

The risk prediction model for complications after flap repair has certain predictive value, but the overall risk of bias is high, and there is a lack of external validation; thus, it needs to be further enhanced and optimized to increase its prediction accuracy and clinical practicability.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12893-025-03072-8.

Introduction

Flap repair surgery is one of the most common microsurgical procedures. It involves transferring a flap of tissue from the donor area to the recipient area in need of repair. Through microsurgical techniques, the blood vessels of the donor and recipient areas are anastomosed to reestablish circulation, thereby repairing the defect, restoring tissue function, and improving appearance. Flap repair plays a crucial role in reconstructive surgery, with widespread applications in trauma repair, reconstruction after tumor resection, and correction of various congenital deformities [1, 2]. Despite the maturity of this technique, postoperative complications such as flap vascular crisis, hematoma, necrosis, and infection occur at relatively high rates, ranging from 9.98–19.6% [3]. These complications remain key factors influencing surgical success rates and patient prognosis. They can not only lead to surgical failure, increase patient suffering and economic burden but also prolong hospitalization, hinder recovery, and affect patients’ quality of life [4].

In recent years, many scholars have developed risk prediction models for flap-related complications to identify high-risk patients preoperatively or early postoperatively to take targeted preventive measures. However, comprehensive comparative studies on the methods used for model construction, the included predictive variables, model performance, and data sample bias are lacking. The predictive capability and clinical applicability of these models remain controversial. This study aims to systematically evaluate the research on postoperative complication risk prediction models for flap repair surgery; assess their accuracy, stability, and clinical value; explore potential factors influencing model performance; and provide guidance for the development and improvement of future models.

Methods

Registration

This systematic review and meta-analysis was reported and conducted in accordance with the PRISMA (Preferred Reporting Items for Systematic Review and Meta-Analysis) statement [5]. The study protocol was registered on PROSPERO (International Prospective Register of Systematic Reviews) under the registration number CRD42024583165.

Literature search strategy

Computerized searches were performed in the PubMed, Embase, Web of Science, Cochrane Library, CNKI, SinoMed, VIP and Wanfang databases for studies published on risk prediction models for flap complications. The search strategy consisted of medical subject headings (Mesh), including surgical flaps, postoperative complications, and a risk prediction model. For example, the search strategy in PubMed was as follows: (Surgical Flaps [Mesh] OR Flap, Surgical [Title/Abstract] OR Flaps, Surgical [Title/Abstract] OR Surgical Flap [Title/Abstract] OR Island Flaps [Title/Abstract] OR Island Flap [Title/Abstract] OR Flap, Island [Title/Abstract] OR Flaps, Island [Title/Abstract] OR Pedicled Flap [Title/Abstract] OR Flap, Pedicled [Title/Abstract] OR Flaps, Pedicled [Title/Abstract] OR Pedicled Flaps [Title/Abstract]) AND (Postoperative Complications [Mesh] OR Complication, Postoperative [Title/Abstract] OR Complications, Postoperative [Title/Abstract] OR Postoperative Complications [Title/Abstract] OR graft failure [Title/Abstract] OR engraftment failure [Title/Abstract] OR failed engraftment [Title/Abstract] OR failed graft [Title/Abstract] OR flap failure [Title/Abstract] OR flap loss [Title/Abstract] OR graft loss [Title/Abstract]) AND (risk prediction model [Title/Abstract] OR risk prediction [Title/Abstract] OR prediction model [Title/Abstract] OR prediction tool [Title/Abstract] OR risk score [Title/Abstract] OR risk assessment [Title/Abstract] OR roc curve [Title/Abstract] OR AUC [Title/Abstract] OR area under curve [Title/Abstract] OR nomogram [Title/Abstract]). The search period was from database inception to December 28, 2024.

Selection criteria

The inclusion criteria were as follows: (1) Study type: cohort studies, case‒control studies, and cross-sectional studies. (2) Study subjects: patients aged ≥ 18 years who have undergone flap repair surgery. (3) Study content: development or validation of risk prediction models for postoperative complications following flap repair. (4) Outcome measures: postoperative complications occurring after flap repair surgery.

The exclusion criteria were as follows: (1) review articles, case reports, or conference abstracts; (2) articles that do not provide valid data; and (3) articles that analyzed only the risk factors for flap complications without constructing a prediction model.

Study selection and data extraction

Two researchers independently screened the literature on the basis of the predefined inclusion and exclusion criteria, excluding studies that did not meet the standards. According to the CHARMS checklist (Checklists for Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modeling Studies) used for systematic review data extraction of risk prediction models, the following data were extracted from the included studies: first author, publication year, country, study design, study subjects, sample size, modeling methods, model performance, model presentation format, predictive factors and their quantity, etc. Data extraction was conducted independently by two researchers. After data extraction, the researchers cross-checked the extracted data to ensure consistency. Any discrepancies in data extraction were resolved through discussion or by consulting a third party. The extracted data was then organized and managed for subsequent analysis.

Model quality assessment

Two researchers used the Prediction Model Risk of Bias Assessment Tool (PROBAST) to assess the risk of bias and applicability of the studies included in the literature [6]. Any disagreements were resolved by consulting a third party. The risk of bias assessment covered four domains—study participants, predictors, outcomes, and data analysis—while the applicability assessment covered three domains—study participants, predictors, and outcomes. Each domain was rated into one of three categories: low risk/high applicability, high risk/low applicability, or unclear.

Statistical analysis

Statistical analysis of the incidence and predictive factors of postoperative complications Following flap repair, the included studies were conducted using Stata 18.0 software. The effect size for the incidence was expressed as a rate (95% CI), whereas the effect size for the predictive factors was represented by the odds ratio (OR) (95% CI). If P > 0.1 and I² ≤ 50%, the heterogeneity between studies was considered nonsignificant, and a fixed-effects model was used for analysis. If P ≤ 0.1 or I² >50%, significant heterogeneity between studies was indicated, and a random effects model was employed. Sensitivity analysis was performed to exclude low-quality studies or fit different models to assess the robustness of the results.

Results

Literature screening results

A total of 847 relevant articles were initially screened from various databases, which included 55 Chinese articles and 792 English articles (Fig. 1). After excluding 98 duplicate articles, 716 articles were further excluded based on a review of their titles and abstracts. This left a total of 33 articles for full-text screening. Among them, 9 articles were excluded as they were unrelated to the topic, 4 because they were unable to extract data, and 4 because they only discussed risk factors without constructing models. Ultimately, 16 articles were included.

Fig. 1.

Literature screening flowchart

Basic characteristics of the included studies

A total of 16 studies were included. The sample sizes of the included studies ranged from 90 to 6475. The number of positive events ranged from 12 to 1,096 cases across 15 studies [7–21]. All were published in the past three years. Among them, 13 studies were from China [7, 8, 10–16, 18–21], one was from Canada [22], one was from the United States [17], and one was conducted by researchers from 11 countries [9]. Twelve studies were single-center studies [7, 8, 10–13, 15, 16, 18–21], and four were multicenter studies [9, 14, 17, 22]. Ten studies were retrospective case‒control studies [8, 11–13, 15, 16, 18–21], five were retrospective cohort studies [7, 10, 14, 17, 22], and one was a prospective cohort study [9]. The research subjects were patients who underwent flap repair surgery. The basic characteristics of the included studies are shown in detail in Table 1.

Table 1.

Basic characteristics of the included studies

Study	Country	Study design	Participants	Period	Sample size
					Total	Events	Incidence
O’Neill 2020 [22]	Canada	Retrospective cohort	Flap repair patients	2009.1-2017.1	1012	12	1.10%
Zhao 2021 [21]	China	Retrospective case‒control	Flap repair patients	2017.1-2020.12	126	42	33.33%
Shi 2022 [18]	China	Retrospective case‒control	Flap repair patients	2006.1.1-2020.12.12	946	34	3.60%
Hassan 2023 [17]	USA	Retrospective cohort	Flap repair patients	2018.1-2019.12	649	40	6.20%
Chen 2023 [16]	China	Retrospective case‒control	Flap repair patients	2010.1.1-2023.6.30	192	37	19.27%
Wu 2023 [15]	China	Retrospective case‒control	Flap repair patients	2009.12-2021.12	540	32	5.92%
Yang 2023 [20]	China	Retrospective case‒control	Flap repair patients	2019.1.1-2021.12.31	570	46	8.07%
Liu 2024 [14]	China	Retrospective cohort	Flap repair patients	2000–2020	193	26	13.47%
Jiang 2024 [19]	China	Retrospective case‒control	Flap repair patients	2018.6-2023.6	90	31	34.44%
Song 2024 [13]	China	Retrospective case‒control	Flap repair patients	2017.1-2023.6	575	59	10.26%
Luo 2022 [10]	China	Retrospective cohort	Flap repair patients	2019.1-2020.10	844	405	47.99%
Li 2022 [11]	China	Retrospective case‒control	Flap repair patients	2017.10-2020.9	120	31	25.83%
Bernuth 2024 [9]	11 countries	Prospective cohort	Flap repair patients	2008-2021	6475	1096	16.93%
Zhao 2022 [12]	China	Retrospective case‒control	Flap repair patients	2019.7-2022.3	416	72	17.30%
Zhang 2024 [8]	China	Retrospective case‒control	Flap repair patients	2006.1.1-2020.12.12	769	21	2.73%
Zheng 2024 [7]	China	Retrospective cohort	Flap repair patients	2018.1-2019.12	1786	63	3.52%

Risk prediction model construction

A total of 16 studies reported 28 risk prediction models for complications following flap repair surgery. The sample size for model development ranged from 90 to 6,475 cases, while the sample size for model validation ranged from 30 to 535 cases. In terms of modeling methods, three studies [17, 18, 22], utilized machine learning algorithms; one study [20], combined logistic regression analysis with machine learning algorithms; and the remaining studies employed logistic regression analysis to build their models. Regarding variable selection, eight studies [7, 10, 12–14, 16, 19, 22], used both machine learning and univariate analysis to select variables, whereas eight studies [8, 9, 11, 15, 17, 18, 20, 21], relied solely on univariate analysis. For handling continuous variables, the final models included 3 to 15 predictive factors. Twelve studies [7–9, 11, 12, 15–18, 20–22], maintained the continuity of continuous variables, whereas four studies [10, 13, 14, 19], converted continuous variables into categorical variables. In terms of handling missing data, one study [20], used mode imputation, regression imputation, and mean imputation; two studies [7, 16], employed multiple imputation; two studies [10, 17], used deletion methods, excluding cases with missing data; and the remaining studies did not explicitly report whether missing data were present. Ten models were presented using nomograms, three using regression equations, two using dendrograms, one using an accumulated local effect (ALE) plot, and one using a neural network plot. One study [18] did not specify the presentation format of the model. Detailed information is provided in Tables 2 and 3.

Table 2.

Construction of risk prediction models for flap complications

Study	Modeling method	Variable selection	Sample size (D/I/E)	Missing data handling	Variable type
O’Neill 2020 [22]	DT	Logistic regression	607/405/-	-	Continuous variable
Zhao 2021 [21]	LR	Univariate and multivariate analysis	96/30/-	-	Continuous variable
Shi 2022 [18]	RF, SVM, GB	Univariate and multivariate analysis	473/473/-	-	Continuous variable
Hassan 2023 [17]	KNN, XGB, NNET, MARS, SVM, DT, GLM, RF, VOTE	Univariate and multivariate analysis	555/139/-	Delete	Continuous variable
Chen 2023 [16]	LR	Logistic regression, Stepwise Regression	192/-/-	Multiple imputation	Continuous variable
Wu 2023 [15]	LR	Univariate and multivariate analysis	378/-/162	-	Continuous variable
Yang 2023 [20]	LR, RF, NN	Univariate and multivariate analysis	317/228/-	Mode substitution, regression imputation, mean substitution	Continuous variable
Liu 2024 [14]	LR	Logistic regression, LASSO regression	150/-/43	-	Categorical variables
Jiang 2024 [19]	LR	Logistic regression	90/-/-	-	Categorical variables
Song 2024 [13]	LR	Logistic regression	403/172/-	-	Categorical variables
Luo 2022 [10]	LR	Univariate and multivariate analysis, LASSO regression, Based on clinical knowledge	844/-/-	Delete	Categorical variables
Li 2022 [11]	LR	Univariate and multivariate analysis	120/-/-	-	Continuous variable
Bernuth 2024 [9]	LR	Univariate and multivariate analysis	6475/-/-	-	Continuous variable
Zhao 2022 [12]	LR	Logistic regression, LASSO regression	416/-/-	-	Continuous variable
Zhang 2024 [8]	LR	Univariate and multivariate analysis	769/-/-	-	Continuous variable
Zheng 2024 [7]	LR	Logistic regression	1251/535/-		Continuous variable

DT decision tree, LR logistic regression, RF random forest, SVM support vector machine, KNN k-nearest neighbors, GB gradient boosting, XGB extreme gradient boosting, MARS multivariate adaptive regression splines, GLM generalized linear model, VOTE ensemble voting, NNET single-layer artificial neural network, NN neural network, D model development, I internal validation, E external validation, - = not reported

Table 3.

Final predictors of the risk prediction model and presentation forms

Study	Final predictors (n)	Model presentation
O’Neill 2020 [22]	Age, BMI, Comorbidities, Smoking, Flap reconstruction time, Number of flap reconstructions, Radiotherapy (7)	Dendrogram
Zhao 2021 [21]	Hypothermia, Diabetes, Smoking, Decubitus position on the affected side, Vasodilator use, Pain (6)	Equation
Shi 2022 [18]	Age, BMI, Flap reconstruction time, Smoking, Diabetes, History of surgery, Chemotherapy, Hypertension, Insulin use, Obesity (10)	NR
Hassan 2023 [17]	Diabetes, High BMI, Age, Hypertension, Subpectoral device placement, Nipple-sparing mastectomy, axillary lymph node dissection, Chemotherapy, Postoperative radiotherapy, No ADM use (11)	ALE plot
Chen 2023 [16]	Preoperative osteomyelitis, Flap type, Total intraoperative fluid intake, Excessive intraoperative blood loss (4)	Nomogram
Wu 2023 [15]	Vascular crisis, Infection, Hematoma (3)	Nomogram
Yang 2023 [20]	BMI, Preoperative fibrin, Preoperative hemoglobin, Operation time, Platelet count, Smoking, Diabetes, Number of venous anastomosis, Flap type, Donor site, Recipient site, Preoperative chemoradiotherapy, Intraoperative blood transfusion, Arteriovenous anastomosis, Postoperative infusion volume (15)	Equation, dendrogram, neural network plot
Liu 2024 [14]	Radiotherapy interval, Trismus, Diabetes, Vein anastomosis, ASA classification (5)	Nomogram
Jiang 2024 [19]	Diabetes, Flap reconstruction time, Hematoma, Infection (4)	Equation
Song 2024 [13]	Flap reconstruction time, Unreasonable selection of anastomotic vessels, Infection, Excessive intraoperative blood loss, Infirm postoperative fixation (5)	Nomogram
Luo 2022 [10]	Age, BMI, Liver disease, Chemotherapy, Number of lymph nodes resected ≥14, Detensioning sutures, Hypertension, Diabetes (8)	Nomogram
Li 2022 [11]	Venous anastomosis, Excessive intraoperative blood loss, Smoking, Level of surgery by physician, Improper design of the flap (5)	Nomogram
Bernuth 2024 [9]	Operation time, ASA classification, Surgical specialty, Outpatient setting, Race, Sex (6)	Nomogram
Zhao 2022 [12]	Surgical history, Hypertension, Smoking, Bone invasion, BMI, Defect size (6)	Nomogram
Zhang 2024 [8]	Diabetes, Surgical site, Surgical method including arterial ligation, Wound effusion, Infection (5)	Nomogram
Zheng 2024 [7]	Tissue flap width, D-dimer, Preoperative hemoglobin, Preoperative and postoperative hemoglobin difference, Intravenous anastomosis, Preoperative albumin value, Prothrombin time, Tissue flap surgery history, Surgical site, Flap type (10)	Nomogram

BMI body mass index, ADM adrenal medulla, ASA American Society of Anesthesiologists

Performance of the prediction models

The discriminatory ability of the included models was evaluated using the area under the receiver operating characteristic curve (AUC) or C-statistic (C-index). Five studies [7, 13–15, 22], reported the AUC during model development, with AUC values ranging from 0.780 to 0.947. Fourteen studies [7, 8, 10, 12–22], reported internal validation AUCs that ranged from 0.655 to 0.964. None of the studies reported the external validation AUC. Among the 16 risk prediction models, 14 demonstrated high predictive performance (AUC > 0.7). Four studies [8, 9, 11, 12], reported the C-index, with values ranging from 0.731 to 0.896, all exceeding 0.7. The specificity ranged from 52.0 to 95.0%, the sensitivity ranged from 27.0 to 100%, and the accuracy ranged from 52.0 to 89.0%. Calibration was assessed using the Hosmer‒Lemeshow test (P > 0.05), calibration curves, or decision curve analysis, with 11 studies [7, 8, 10–17, 19], reporting calibration results. Concerning model validation, all 14 studies conducted internal validation, mainly using the bootstrap method and cross-validation. Three studies [7, 14, 15] performed external validation. Detailed information is provided in Table 4.

Table 4.

Performance and presentation formats of risk prediction models for flap complications

Study	Model performance				Model validation
	AUC(D/I/E)	C-index	Specificity/Sensitivity/Accuracy	Calibration method
O’Neill 2020 [22]	0.947/0.672/-	-	87.8%/100%/-	-	Internal, bootstrap
Zhao 2021 [21]	-/0.884/-	-	87.5%/78.57%/83.33%	-	-
Shi 2022 [18]	RF: -/0.770/- SVM: -/0.720/- GB: -/0.707/-	-	RF: 78.0%/-/- SVM: 71.0%/-/- GB: 68.0%/-/-	-	Internal, 5-fold cross- validation
Hassan 2023 [17]	KNN: -/0.67/- XGB: -/0.58/- NNET: -/0.53/- MARS: -/0.68/- SVM: -/0.59/- DT: -/0.51/- GLM: -/0.63/- RF: -/0.700/- VOTE: -/0.61/-	-	KNN: 83.0%/36.0%/79.0% XGB: 52.0%/55.0%/52.0% NNET: 88.0%/27.0%/76.0% MARS: 77.0%/45.0%/75.0% SVM: 93.0%/27.0%/88.0% DT: 80.0%/27.0%/76.0% GLM: 70.0%/45.0%/68.0% RF: 95.0%/27.0%/89.0% VOTE: 95.0%/27.0%/83.0%	DAC	Internal, 10-fold cross- validation
Chen 2023 [16]	-/0.759/-	-	-/-/-	DAC, Calibration curves	Internal, bootstrap
Wu 2023 [15]	0.871/0.791/-	-	84.97%/84.94%/-	Calibration curves	External
Yang 2023 [20]	LR: -/0.775/- RF: -/0.730/- NN: -/0.828/-	-	LR:78.7%/76.2%/78.5% RF:84.4%/57.1%/86.0% NN:84.4%/85.7%/78.1%	-	Internal, 5-fold cross- validation
Liu 2024 [14]	0.936/0.964/-	-	-/-/-	DAC, H-L Test, Calibration curves	Internal, bootstrap;external
Jiang 2024 [19]	-/0.807/-	-	70.53%/85.29%/-	H-L Test	-
Song 2024 [13]	0.809/0.827/-	-	-/-/-	H-L Test	-
Luo 2022 [10]	-/0.665/-	0.675	-/-/-	DAC, H-L Test, Calibration curves	, Internal, bootstrap
Li 2022 [11]	-/-/-	0.829	-/-/-	Calibration curves	Internal, bootstrap
Bernuth 2024 [9]	-/-/-	0.731	-/-/-	-	Internal, bootstrap
Zhao 2022 [12]	-/0.763/-	0.747	-/-/-	DAC	Internal, bootstrap
Zhang 2024 [8]	-/0.900/-	0.896	91.2%/76.2%/-	DAC	-
Zheng 2024 [7]	0.780/0.701/-	-	76.4%/70.8%/-	DAC, H-L Test, Calibration curves	Internal, bootstrap; external

AUC area under the curve, C-index C statistic, DT decision tree, LR logistic regression, RF random forest, SVM support vector machine, KNN k-nearest neighbors, GB gradient boosting, XGB extreme gradient boosting, MARS multivariate adaptive regression splines, GLM generalized linear model, VOTE ensemble voting, NNET single-layer artificial neural network, NN neural network, D model development, I internal validation, E external validation, H-L test Hosmer–Lemeshow test, DAC decision analysis curve, Bootstrap bootstrapping method, - = not reported

Risk of bias and applicability assessment

Two researchers conducted a comprehensive assessment of the literature quality using the PROBAST tool [6], and carefully reviewed the evaluation results to ensure accuracy. In the domain of study participants, 15 studies [7, 8, 10–22], were retrospective studies, which were considered to have a high risk of bias, whereas one study [9], was a prospective cohort study and was deemed to have a low risk of bias. Regarding predictor factors, all studies were assessed as having a low risk of bias. Likewise, in the outcomes domain, all studies exhibited a low risk of bias. In the data analysis domain, three studies [7, 12, 16], had a low risk of bias, while 13 studies [8–11, 13–15, 17–22], were determined to have a high risk of bias. Among these, two studies [19, 21], had an insufficient sample size in the modeling group; the events per variable (EPV) for each predictor variable were < 20; four studies [10, 13, 14, 19] converted continuous variables into categorical variables; eight studies [8, 9, 11, 15, 17, 18, 20, 21], converted continuous variables into categorical variables; eight studies [9, 18, 20–22], did not report whether model calibration tests were performed. All included studies demonstrated good applicability in terms of the number of study participants, predictor factors, and outcomes, as shown in Table 5.

Table 5.

Assessment of the quality of the included studies

Study	ROB				Applicability			Overall
	Participants	Predictors	Outcome	Analysis	Participants	Predictors	Outcome	ROB	Applicability
O’Neill 2020 [22]	-	+	+	-	+	+	+	-	+
Zhao 2021 [21]	-	+	+	-	+	+	+	-	+
Shi 2022 [18]	-	+	+	-	+	+	+	-	+
Hassan 2023 [17]	-	+	+	-	+	+	+	-	+
Chen 2023 [16]	-	+	+	+	+	+	+	-	+
Wu 2023 [15]	-	+	+	-	+	+	+	-	+
Yang 2023 [20]	-	+	+	-	+	+	+	-	+
Liu 2024 [14]	-	+	+	-	+	+	+	-	+
Jiang 2024 [19]	-	+	+	-	+	+	+	-	+
Song 2024 [13]	-	+	+	-	+	+	+	-	+
Luo 2022 [10]	-	+	+	-	+	+	+	-	+
Li 2022 [11]	-	+	+	-	+	+	+	-	+
Bernuth 2024 [9]	+	+	+	-	+	+	+	-	+
Zhao 2022 [12]	-	+	+	+	+	+	+	-	+
Zhang 2024 [8]	-	+	+	-	+	+	+	-	+
Zheng 2024 [7]	-	+	+	+	+	+	+	-	+

"+" indicates a low risk of bias/high applicability; "−" indicates a high risk of bias/low applicability; "?" indicates unclear

Meta-analysis results

A meta-analysis was conducted on the incidence of complications following flap repair surgery based on 14 studies. The heterogeneity test revealed significant heterogeneity (P < 0.001, I² = 99.1%), and a random effects model was used. The results remained stable even after the removal of any single study, indicating the robustness of the findings. The overall incidence of complications after flap repair surgery was 14.8% (95% CI, 10.7–19.0%), as shown in Fig. 2. A meta-analysis was also conducted for predictors, with at least three studies reporting valid odds ratios (ORs). The predictive factors included age, body mass index, smoking history, flap reconstruction time, diabetes, chemotherapy history, hypertension, postoperative infection, and excessive intraoperative blood loss. The results revealed that BMI (OR = 1.15, 95%CI: 1.05–1.26, I² = 61.4%), smoking history (OR = 2.70, 95%CI: 1.93–3.77, I² = 0%), prolonged flap reconstruction time (OR = 2.46, 95%CI: 1.98–3.06, I² = 42.1%), diabetes (OR = 2.27, 95%CI: 1.48–3.49, I² = 59.8%), hypertension (OR = 1.71, 95%CI: 1.30–2.26, I² = 46.6%), and postoperative infection (OR = 2.91, 95%CI: 2.22–3.82, I² = 32.8%) were independent risk factors for complications after flap repair surgery (P < 0.05). The results are presented in Table 6.

Fig. 2.

Forest plot of the pooled incidence rates of postoperative complications following flap reconstruction

Table 6.

Meta-analysis of factors influencing postoperative complications following flap reconstruction

Predictors	Effect model	Heterogeneity test		Pooled effect size
		I2(%)	P	OR (95% CI)	Z	P
Age	Random-effects model	69.2	0.021	1.25 (0.88, 1.78)	1.262	0.207
BMI	Random-effects model	61.4	0.024	1.15 (1.05, 1.26)	2.888	0.004
Smoking	Fixed-effect model	0	0.643	2.70 (1.93, 3.77)	5.832	< 0.001
Flap reconstruction time	Fixed-effect model	42.1	0.159	2.46 (1.98, 3.06)	8.105	< 0.001
Diabetes	Random-effects model	59.8	0.015	2.27 (1.48, 3.49)	3.743	< 0.001
Chemotherapy	Random-effects model	86.1	< 0.001	0.92 (0.34, 2.49)	−0.174	0.862
Hypertension	Fixed-effect model	46.6	0.132	1.71 (1.30, 2.26)	3.775	< 0.001
Postoperative infection	Fixed-effect model	32.8	0.216	2.91 (2.22, 3.82)	7.698	< 0.001
Excessive blood loss during surgery	Random-effects model	78.9	0.009	1.59 (0.48, 5.26)	0.761	0.447

Discussion

This study thoroughly integrated and analyzed prediction models for postoperative flap complications, encompassing 16 studies and a total of 28 models. Most flap complication prediction models demonstrated high discrimination, with 16 models exhibiting an AUC > 0.7 and 4 models showing a C-index > 0.7. These findings suggest that the majority of the models displayed good predictive performance and effectively identified high-risk patients for complications following flap reconstruction. Among them, the model developed by Liu et al. [14] achieved the highest AUC of 0.964, indicating that the model could accurately rank patients with and without complications 96.4% of the time. This can assist clinicians in early identification of high-risk individuals and in implementing targeted preventive measures. Although most models performed well in terms of prediction, the overall risk of bias was high. The primary reasons for this and the aspects to consider in future modeling are as follows. Firstly, most studies were retrospective in design, which may introduce data bias and issues with missing data. Future studies should aim to collect data prospectively to reduce these sources of bias in the models. Secondly, half of the studies used univariate analysis to select predictive factors. Relying on a single factor may reduce selection accuracy and increase bias. Future variable selection should combine clinical practice with new methods to improve selection accuracy [23]. Thirdly, most studies did not report methods for handling missing data. It is advisable to use reasonable methods, such as imputation or filling techniques, to manage missing values and reduce the risks of bias and overfitting [24]. Further, some studies converted continuous variables into binary categories. It is recommended to maintain the continuity of variables as much as possible to ensure model stability, avoid information loss, ensure accurate parameter estimation, and improve generalization ability. Lastly, most studies did not perform external validation, making the generalizability of the models uncertain. Model development should include both internal and external validation to optimize in-sample performance, correct biases, and test applicability in different settings. This helps improve overall performance, reduce bias, and ensure the model’s utility. In summary, risk prediction models for postoperative flap complications are still in the exploratory stage. Future researchers can use the PROBAST evaluation tool to develop scientifically sound and comprehensive research plans to improve the performance of prediction models.

Early identification and management of high-risk patients for postoperative flap complications are crucial for enhancing flap survival rates. The results of this study indicate that BMI, smoking history, prolonged flap reconstruction time, comorbid diabetes, comorbid hypertension, and postoperative infection are independent risk factors for postoperative flap complications. Patients with a high BMI typically have more adipose tissue, which has a relatively poor blood supply. In flap surgery, this can interfere with the blood perfusion of the flap [25]. For high-BMI patients, preoperative and postoperative vascular ultrasound examinations should be performed to assess the vascular condition. For patients with a smoking history, once nicotine enters the bloodstream, it stimulates the sympathetic nervous system, leading to increased vasoconstriction and reduced blood circulation to the flap. It can also increase serum nitric oxide levels, which, through their effect on oxygen, lead to a lower oxygen content in the blood, affecting the oxygen supply to the transplanted flap [26]. Preoperative smoking cessation counseling is a critical step. For patients with a smoking history, enhanced flap monitoring should be implemented postoperatively. During reconstruction, flaps are subjected to ischemic conditions for extended periods. When blood supply is reestablished, a series of complex pathophysiological responses, including the generation of reactive oxygen species, inflammatory responses, Ca²⁺ overload, and apoptosis, are triggered [27]. Studies have analyzed the relationship between reconstruction time and flap necrosis, indicating that the longer the reconstruction time, the higher the rate of flap necrosis [28]. Surgeons should plan the surgery carefully in advance to enhance team efficiency and reduce the reconstruction time. Patients with diabetes are in a prolonged hyperglycemic state, which can cause microvascular and macrovascular damage, leading to increased blood viscosity and affecting the blood supply around the flap postoperatively [29]. Additionally, the immune function of diabetic patients is impaired, reducing their ability to fight infections and increasing the risk of flap infection [30]. Strict blood glucose control should be enforced preoperatively, and postoperative flap circulation should be closely monitored. Patients with hypertension may experience vascular damage and spasm, and they are also at risk of excessive intraoperative bleeding, which raises the incidence of complications [31]. Strict control of the patient’s blood pressure and enhanced intraoperative blood pressure monitoring are critical. Postoperative infection can obstruct blood circulation in the affected area through mechanisms such as inflammatory responses and bacterial toxin damage [32]. Further research is needed to develop more precise methods for infection prevention and treatment.

In clinical practice, early identification and prevention of postoperative flap complications are critical. To advance model development in this field, future research should leverage big data analysis and machine learning technologies to identify the most predictive key variables and build more streamlined and efficient models. In addition, it is essential to establish multicenter, large-sample collaborative studies by standardizing research protocols and methodologies to reduce heterogeneity, thereby enhancing the reliability and generalizability of the models. Furthermore, a strong focus should be placed on external validation by testing the model’s effectiveness in various clinical scenarios and patient populations to ensure its broad applicability in real-world settings. Finally, new influencing factors should be incorporated, as the continuous advancement of medical technology and the accumulation of clinical experience may lead to the identification of additional relevant variables, which should be integrated into the development of future models.

This study has several limitations that should be acknowledged. First, the majority of the included studies were retrospective in nature, which may introduce inherent biases and limit the strength of the evidence. Additionally, some studies reported only selected performance metrics, restricting a comprehensive assessment of the predictive models. Furthermore, external validation was largely absent across most studies, raising concerns about the generalizability and real-world applicability of the proposed models. These limitations highlight the need for more rigorously designed prospective studies and broader validation efforts to enhance the reliability and utility of predictive models in clinical settings.

Conclusion

This study systematically evaluated 16 studies. The results indicate that the risk prediction models for complications after flap repair surgery generally have significant clinical application potential but still require further improvement and optimization. However, all studies exhibited a high risk of bias, and most models lacked external validation. A meta-analysis was performed on the predictive factors. BMI, smoking history, long flap reconstruction time, diabetes mellitus, hypertension, and postoperative infection are identified as risk factors for complications and should be prioritized in clinical work. Future research should emphasize optimizing modeling methods, strengthening external validation, ensuring comprehensiveness and standardization of predictive factors, and developing more accurate risk prediction models by incorporating multicenter and diverse clinical data to provide more effective tools for clinical practice.

Authors’ contributions

Yang Jiebin was responsible for the research design and writing of the manuscript. Qin Xinya and Liu Yamei were in charge of data collection and statistical analysis. Hou Lili is responsible for manuscript editing and review. All the authors reviewed the manuscript.

Funding

This study was funded by the Construction of the Advanced Specialized Nursing Base of the School of Nursing, Shanghai Jiao Tong University. Project Number: (Hlgy1901sjk).

Data availability

Data is provided within the manuscript or supplementary information files.

Declarations

Ethics approval and consent to participate

Not applicable. This study does not involve human participants.

Consent for publication

Not applicable. This study does not involve animal or human participants.

Competing interests

The authors declare no competing interests.