Introduction

Breast cancer is the most common cancer among women in 154 countries and the main cause of cancer-related death in 103 countries. In 2018, there were approximately 2.1 million new cases of breast cancer in women, accounting for 24.2% of the total cases, and the mortality rate was approximately 15.0% [1].

Survival is defined as the period of time a patient survives after disease diagnosis.The 5-year threshold is important to standardize reporting and to identify survivability. Labelling a patient record as survived or not survived takes at least 5 years, therefore, some previous studies used a 5-year threshold to identify the cohort’s survivability [2]. Breast cancer is a complex disease, and although its survival rates in recent years have increased gradually, its 5-year survival rate is considerably different between individuals [3]. Predicting breast cancer survival accurately could help doctors make better decisions regarding medical treatment intervention planning, prevent excessive treatment, thereby reducing economic costs [4, 5], more effectively include and exclude patients in a randomized trial [6], and develop palliative care and hospice care systems [7, 8]. Therefore, predicting survival has become a major issue in current research on breast cancer.

With the surge of medical data as well as the rapid development of information technology and artificial intelligence, the application of big data analysis technology in the construction of survival prediction model has become a current research hotspot. Traditional prediction models based on prior hypothesized knowledge often consider the relationship between dependent variables; in contrast, ML has the potential of learning data models automatically, does not require any implicit assumptions and is able to handle interdependence and nonlinear relationships between variables [9]. It has natural strengths in dealing with the very large number of complex higher-order interactions of medical data. Therefore ML tools have a high potential for application in routine medical practice as leading tools in health informatics.

A growing number of ML studies have been applied to diagnosis [10–13], disease risk prediction [14], recurrence prediction [15], and symptom prediction [16–19]. Furthermore, although the number of survival predictions increases gradually, the database set, modeling process, methodological quality, performance metrics, and modeling of related candidate predictors exhibit large differences [20].

This article aims to systematically and comprehensively review the published literature regarding the use of ML algorithms for model development and validation of breast cancer survival prediction. The primary outcome indicator is the accuracy of the different models in predicting 5-year (60 months or 1825 days) survival rate for breast cancer with the goal of providing a better theoretical basis for the application of ML in survival prediction.

Methods

Results

Search results

By searching three medical databases, a total of 8193 studies were identified. After removing duplicates studies. there were leaving 2829 studies and 2656 studies were eliminated based on the screening of titles and abstracts. A comprehensive review of the full text of the remaining 173 studies was conducted, and 142 were excluded for the following reasons: the type of literature did not meet the criteria, i.e., conference abstracts, books, and review literature (n = 9); the predictive outcome was not 5-year survival but recurrence, survival status, benign and malignant tumor diagnosis, or treatment symptoms (n = 91); the full text was unavailable (n = 6); the data were incomplete (n = 14); the study was not published in English (n = 1); or the study included animal research (n = 1). A total of 31 studies met the inclusion criteria [2, 23–52]. The literature screening process is shown in Fig 1.

Fig 1

PRISMA flowchart.

Assessment of the risk of bias

Among the 31 studies, 9 had a high risk of bias [2, 25, 27, 28, 43, 44, 46, 48, 50], 17 had a moderate risk of bias [24, 26, 29–35, 39, 41, 42, 45, 47, 49, 51, 52], and 5 ad a low risk of bias [23, 36–38, 40], as shown in Table 1.

Table 1

Risk of bias and applicability assessment grading of 31 studies as per the PROBAST criteria.

Research	Participant bias	Predictor bias	Outcome bias	Analysis bias	Overall bias rating	Overall applicability rating
Delen, 2005 [2]	Low	Low	Low	High	High	Low
Bellaachia, 2006 [23]	Low	Low	Low	Low	Low	Low
Endo, 2007 [24]	Low	Low	Low	Moderate	Moderate	Low
Khan, 2008 [25]	Low	Low	Low	High	High	Low
Thongkam, 2008 [26]	Low	Low	Low	Moderate	Moderate	Low
Choi, 2009 [27]	Low	Low	Low	High	High	Low
Liu, 2009 [28]	Low	Low	Low	High	High	Low
Wang, 2013 [29]	Low	Low	Low	Moderate	Moderate	Low
Kim, 2013 [30]	Low	Low	Low	Moderate	Moderate	Low
Park, 2013 [31]	Low	Low	Low	Moderate	Moderate	Low
Shin, 2014 [32]	Low	Low	Low	Moderate	Moderate	Low
Wang, 2015 [33]	Low	Low	Low	Moderate	Moderate	Low
Wang, 2014 [34]	Low	Low	Low	Moderate	Moderate	Low
Chao, 2014 [35]	Low	Low	Low	Moderate	Moderate	Low
García-Laencina, 2015 [36]	Low	Low	Low	Low	Low	Low
Lotfnezhad Afshar, 2015 [37]	Low	Low	Low	Low	Low	Low
Khalkhali, 2016 [38]	Low	Low	Low	Low	Low	Low
Shawky, 2016 [39]	Low	Low	Low	Moderate	Moderate	Low
Sun, 2018 [40]	Low	Low	Low	Low	Low	Low
Sun, 2018 [41]	Low	Low	Low	Moderate	Moderate	Low
Zhao, 2018 [42]	Low	Low	Low	Moderate	Moderate	Low
Fu, 2018 [43]	Low	Low	High	High	High	High
Lu, 2019 [44]	Low	Low	Low	High	High	Low
Abdikenov, 2019 [45]	Low	Low	Low	Moderate	Moderate	Low
Kalafi, 2019 [46]	Low	Low	Low	High	High	Low
Shouket, 2019 [47]	Low	Low	Low	Moderate	Moderate	Low
Ganggayah, 2019 [48]	Low	Low	Low	High	High	Low
Simsek, 2020 [49]	Low	Low	Low	Moderate	Moderate	Low
Salehi, 2020 [50]	Low	Low	Low	High	High	Low
Tang, 2020 [51]	Low	Moderate	Low	Moderate	Moderate	Moderate
Hussain, 2020 [52]	Low	Low	Low	Moderate	Moderate	Low

Primary characteristics of the literature

The primary characteristics of the 31 studies are shown in Table 2. Most of the 31 studies were published from 2013 to 2020, and the statistics regarding the publication year and number of studies are shown in Fig 2. Among them, 22 studies were located in Asia [24, 25, 27–35, 37, 38, 40, 41, 43, 46–48, 50–52], 5 in North America [2, 23, 42, 44, 49], 2 in Oceania [26, 45], 1 in Europe [36], and 1 in Africa [39]. The primary prediction outcome was the 5-year survival of breast cancer patients. The predicted disease types were all breast cancer rather than one particular subtype (e.g., triple-negative breast cancer). All included studies focused on the development of survival prediction models using ML algorithms rather than validating the existing models on independent data. Primary characteristics are shown in Tables 2 and 3.

Fig 2

Number of studies published each year.

Table 2

Primary characteristics and details of the 31 studies (Table 2 was uploaded as an additional file).

First author, year	Country of research	Source of data	Number of samples (Alive+ Dead)	Missingness processing described	Preprocessing described	Feature selection described	Number of candidate predictors used	ML algorithms	Validation method	Model evaluation metrics	Hyperparameter tuning
Delen, 2005 [2]	USA	SEER (1973–2000)	202932 (109659+93273)	Yes	Yes	No	16	ANN; C5 DT; LR	Internal validation	Accuracy; Sensitivity; Specificity	No
Bellaachia, 2006 [23]	USA	SEER (1973–2002)	151886 (116738+35148)	No	Yes	No	16	Naive Bayes; MLP; C4.5 DT	Internal validation	Accuracy; Sensitivity (Recall); Precision	No
Endo, 2008 [24]	Japan	SEER (1972–1997)	37256 (30734+6882)	No	Yes	No	10	LR; ANN; Naive Bayes; Bayesian network; DT + Naive Bayes; DT ID3; DT J48	Internal validation	Accuracy; Sensitivity; Specificity	No
Khan, 2008 [25]	Korea	SEER (1973–2003)	162500 (unknown)	Yes	Yes	No	16	Fuzzy decision trees	Internal validation	Accuracy; Sensitivity; Specificity	No
Thongkam, 2008 [26]	Australia	Srinagarind Hospital records in Thailand (1990–2001)	732 (342+394)	No	Yes	No	11	C4.5 DTs; Conjunctive rule; Naive Bayes; NN-classifier; random committee; RF; Radial basis function network	Internal validation	Accuracy; AUC	No
Choi, 2009 [27]	Korea	SEER (1973–2003)	294275 (unknown)	Yes	Yes	No	14	ANN; Bayesian network; Hybrid Bayesian network	Internal validation	Accuracy; Sensitivity; Specificity; AUC	No
Liu, 2009 [28]	China	SEER (1973–2004)	182517 (157916+24601)	Yes	Yes	Yes	16	C5 DT	Internal validation	Accuracy; Sensitivity; Specificity; AUC	No
Wang, 2013 [29]	China	SEER (1973–2007)	215221 (195172+20049)	Yes	Yes	Yes	9	DT; LR	Internal validation	Accuracy; Sensitivity; Specificity; AUC; G-mean	No
Kim, 2013 [30]	Korea	SEER (1973–2003)	162500 (128469+34031)	No	Yes	No	16	SVM; ANN; Semi-supervised learning; Semi-supervised learning-Co training	Internal validation	Accuracy; AUC	No
Park, 2013 [31]	Korea	SEER (1973–2003)	162500 (128469+34031)	No	Yes	No	16	SVM; ANN; Semi-supervised learning	Internal validation	Accuracy; Sensitivity; Specificity; AUC	Yes
Shin, 2014 [32]	Korea	SEER (1973–2003)	162500 (128469+34031)	No	Yes	No	16	DT; ANN; SVM; Semi-supervised learning; Semi-supervised learning-Co training	Internal validation	AUC	Yes
Wang, 2015 [33]	China	Northern Taiwan hospital database (2000–2003)	604 (464+140)	Yes	Yes	Yes	5	ANN	Internal validation	Accuracy; Sensitivity; Specificity; AUC	No
Wang, 2014 [34]	China	SEER (1973–2007)	215221 (195172+20049)	No	Yes	No	20	LR; C5 DT; 1-nearest neighbor	Internal validation	Accuracy; Sensitivity; Specificity; G-mean	Yes
Chao, 2014 [35]	China	Breast cancer incidence database in Taiwan hospital (2002–2010)	1340 (1272+68)	No	Yes	No	7	SVM; LR; C5 DT	Internal validation	Accuracy	Yes
García-Laencina, 2015 [36]	Spain	Institute Portuguese of Oncology of Porto	399 (282+117)	Yes	Yes	No	16	KNN; Classification Trees; LR; SVM	Internal validation	Accuracy; Sensitivity; Specificity; AUC	Yes 2-D grid search
Lotfnezhad Afshar, 2015 [37]	Iran	SEER (1999–2004)	22763 (Aliveness values were approximately nine times greater than death values)	Yes	Yes	No	18	SVM; Bayesian network; Chi-squared Automatic Interaction Detection	Internal validation	Accuracy; Sensitivity; Specificity; Adjusted propensity	No
Khalkhali, 2016 [38]	Iran	Omid Treatment and Research Center database (2007–2010)	569 (unknown)	Yes	Yes	No	15	Classification and regression tree	Internal validation	Accuracy; Sensitivity; Specificity	No
Shawky, 2017 [39]	Egypt	SEER (2010)	4490 (2245+2245)	Yes	Yes	No	14	ANN; KNN; SVM; LR	Internal validation	Accuracy; Sensitivity; Specificity; AUC	No
Sun, 2018 [40]	China	Molecular Taxonomy of Breast Cancer International Consortium	1980 (1489+491)	Yes	Yes	Yes	3 types	SVM; RF; LR; DNN	Internal and external validation	Accuracy; Sensitivity; Precision; Specificity; AUC; Mcc	Yes
Sun, 2018 [41]	China	The Cancer Genome Atlas	578 (133+445)	No	Yes	Yes	5 types	Multiple kernel learning; Regular Cox models; Parametric censored regression models; Random survival forests; Boosting concordance index; Supervised principal components regression	Internal validation	Accuracy; Sensitivity; Specificity; Precision; AUC; Mcc; C-index	No
Zhao, 2018 [42]	USA	Molecular Taxonomy of Breast Cancer International Consortium	1874 (1409+465)	Yes	Yes	No	27	Gradient Boosting; RF; SVM; ANN	Internal validation	Accuracy; AUC	No
Fu, 2018 [43]	China	The Clinical Research Center for Breast in West China Hospital of Sichuan University (1989–2007)	5246 (1181+4065)	Yes	Yes	Yes	23	Gradient Boosting; DT framework; SVM; RF; Adaboost; Cox Regression	Internal validation	Cutoff; Youden index; Sensitivity; F-score; Specificity; AUC	Yes TPE algorithm
Lu, 2019 [44]	USA	SEER (1973–2014)	82707 (76716+5991)	Yes	Yes	No	14	Genetic algorithm-based online gradient; Online Sequential Extreme Learning Machine; Online Adaptive Boosting with the Adaptive Linear Regressor; Online Gradient Boosting with the Adaptive Linear Regressor; Online linear regressor; AdaBoost; SVM; MLP	Internal validation	Accuracy; Sensitivity; Specificity; AUC; Retraining time	Yes
Abdikenov, 2019 [45]	Australia	SEER (2004–2014)	659802 (376087+283715)	No	Yes	No	19	DNN; LR; SVM; RF; Gradient Boosting	Internal validation	Accuracy; Sensitivity; Specificity; AUC; F1 score	Yes NSGAIII
Kalafi, 2019 [46]	Malaysia	The University Malaya Medical Centre Breast Cancer Registry database (1993–2017)	4902 (2451+2451)	Yes	Yes	No	23	SVM; RF; DT; MLP	Internal validation	Accuracy; Sensitivity; Precision; Specificity; F1 score; Mcc; NPV; FPR; FDR; FNR	No
Shouket, 2019 [47]	Pakistan	Institute of Nuclear Medicine & Oncology Lahore Hospital of Pakistan database (2013–2018)	200 (5-year survival: 190+10; 5-year disease free survival: 164+36)	No	Yes	No	10	Naive Bayes; DT J48; SVM; RF; AdaBoost; JRip	Internal validation	Accuracy; Precision; AUC; F-score; Mcc; NPV	No
Ganggayah, 2019 [48]	Malaysia	The University Malaya Medical Centre Breast Cancer Registry database (1993–2017)	8066 (5614+2452)	Yes	Yes	Yes	23	DT (rpart); RF; ANN; XGboost; LR; SVM	Internal validation	Accuracy; Sensitivity; Specificity; AUC; Precision; Mcc; Calibration curve	No
Simsek, 2020 [49]	USA	SEER (1973–2013)	53732 (1-year: 52886+886; 5-year: 46724+7028; 10-year: 42965+10787.)	Yes	Yes	Yes	17	ANN; LR	Internal validation	Accuracy; Sensitivity; Specificity; AUC	No
Salehi, 2020 [50]	Iran	SEER (2004–2013)	141254 (118324+22930)	Yes	Yes	No	35	MLP; MLP experts; MLP stacked generalization	Internal validation	Accuracy; Sensitivity; Specificity	No
Tang, 2020 [51]	China	Haberman’s Survival Data Set (1958–1970)	306 (unknown)	No	Yes	No	3	Evolutionary dendritic neuron model; MLP; MLP with adaptive learning rate and momentum coefficient; DT; SVM trained by the radial basis function kernel; SVM trained by the linear kernel; SVM trained by the polynomial kernel; Dendritic neuron model	Internal validation	Accuracy; Sensitivity; Specificity; AUC; F-score	Yes
Hussain, 2020 [52]	Iraq	SEER (1973–2001)	90308 (unknown)	Yes	Yes	No	17	ANN; DT; LR	Internal validation	Accuracy; Sensitivity; Specificity	No

Table 3

Primary characteristics and categories of the 31 studies.

Characteristics	Categories	Number (n)	Percentage (%)
Place of research	Asia	22	71.0
	North America	5	16.1
	Oceania	2	6.5
	Europe	1	3.2
	Africa	1	3.2
Published type	Journal article	27	87.1
	Conference paper	3	9.6
	Information paper	1	3.2
Source of data	SEER	18	58.1
	Molecular Taxonomy of Breast Cancer International Consortium	2	6.5
	The Cancer Genome Atlas	1	3.2
	Haberman’s Cancer Survival Dataset	1	3.2
	Hospital Registration Data	9	29.0
Type of data	Public	22	71.0
	Private	9	29.0
Number of centers	Single center	22	71.0
	Multiple centers	9	29.0
Sample size	<1000	7	22.6
	1000~10000	7	22.6
	>10000	17	54.8
Missing data and processing described	Yes	20	64.5
	No	11	35.3
Preprocessing described	Yes	31	100.0
	No	0	0.0
Feature selection described	Yes	8	25.8
	No	23	74.2
Class imbalance processing	Yes	24	77.4
	No	2	6.5
	Unknown	5	16.1
Number of candidate predictors	<10	4	12.9
	10~100	25	80.6
	>100	2	6.5
Number of ML algorithms	1	5	16.1
	>1	26	83.9
Type of ML algorithms	DT	19	61.3
	ANN	18	58.1
	SVM	16	51.6
	LR	12	38.7
	Bayesian classification algorithms	6	19.4
	KNN	3	9.7
	Semi-supervised learning	3	9.7
	Ensemble learning	10	32.3
	DNN	3	9.7
Model presentation	Formula	6	19.4
	Graph	5	16.1
	Formula and graph	16	51.6
	No presentation	4	12.9
Calibration	Yes	1	3.2
	No	30	96.8
Internal validation	Yes	31	100.0
	No	0	0.0
External validation	Yes	1	3.2
	No	30	96.8
Hyperparameter selection	Yes	9	29.0
	No	22	71.0
Model evaluation metrics	Accuracy	29	93.5
	Sensitivity/Recall	25	80.6
	Specificity	24	77.4
	AUC	20	64.5
	Precision/Positive predictive value	6	19.4
	F1 score	5	16.1
	Mcc	5	16.1
	NPV	2	6.5
	G-mean	2	6.5
	C-index	1	3.2
	Cutoff	1	3.2
	Youden index	1	3.2
	Retaining time	1	3.2
	FPR	1	3.2
	FDR	1	3.2
	FNR	1	3.2
Type of candidate predictors	Clinical data	29	93.5
	Clinical data + molecular data	1	3.2
	Clinical data + molecular data + pathological images	1	3.2
Ranking of candidate predictors	Yes	15	28.4
	No	16	51.6

Candidate predictors

The median number of candidate predictors used was 16 (range: 3~625); 29 studies used only clinical data [2, 23–39, 42–52], 1 study combined clinical data with molecular data for prediction [40], and 1 study combined clinical data, molecular data and pathological image data for prediction [41]. We ranked the frequency of the use of certain predictors from high to low. The commonly used candidate predictors included age, stage of cancer, grade, tumor size, race, marital status, number of nodes, histology, number of positive nodes, primary site code, extension of tumor, behavior/behavior code, lymph node involvement, site-specific surgery code, number of primaries, radiation, received radiation, estrogen receptor (ER) status, and progesterone receptor (PR) status (see Table 4).

Table 4

Rank of the candidate predictors used in 31 studies.

Rank	Candidate predictor	Description	Number (n)	Percentage (%)
1	Age	Age at diagnosis	26	83.9
2	Stage of cancer	Defined by size of cancer tumor and its spread	23	74.2
3	Grade	Appearance of tumors and their differentiability	22	71.0
4	Tumor size	Diameter of tumor	21	67.7
5	Race	Recoded race of the patient. Ethnicity: White, Black, Chinese, etc.	19	61.3
6	Marital status	Patient’s marital status at the time of diagnosis: Married, single, divorced, widowed, separated	19	61.3
7	Number of nodes	Total nodes (positive/negative) examined	19	61.3
8	Histology	The microscopic composition of cells and/or tissue for a specific primary	18	58.1
9	Number of positive nodes	When lymph nodes are involved in cancer, they are called positive	18	58.1
10	Primary site code	Presence of tumor at particular location in body. Topographical classification of cancer	17	54.8
11	Extension of tumor	Defines spread of tumor relative to breast	17	54.8
12	Behavior/behavior code	In situ or malignant	15	48.4
13	Lymph node involvement	None, minimal, significant, etc.	14	45.2
14	Site-specific surgery code	Information on surgery during first course of therapy, whether cancer-directed or not	13	41.9
15	Number of primaries	Number of primary tumors	13	41.9
16	Radiation	None, beam radiation, radioisotopes, refused, recommended, etc.	11	35.5
17	Received radiation	Whether the patient had been treated with radiotherapy or not	8	25.8
18	Estrogen receptor (ER) status	Breast cancers with this hormone receptor are called “ER positive”	7	22.6
19	Progesterone receptor (PR) status	Breast cancers with this hormone receptor are called “PR positive”	7	22.6

Fifteen studies ranked the degree to which the predictors contributed to the outcome [2, 23, 25, 27, 32, 33, 37, 38, 42, 43, 46, 48–50, 52]. Four studies reported sequencing methods, including sensitivity analysis in networks [2, 27], DT information gain measurement [23, 25], sensitivity scores in rules [38], and correlation coefficients [33] (see S6 Table).

Discussion

To the best of our knowledge, this is the first systematic review of the application of ML to breast cancer survival prediction, and accurate 5-year survival predictions are very important for further research. After a systematic analysis of 31 studies, we found that there is a need for the standardization and validation of the different algorithms of models for predicting breast cancer survival and for the exploration of the significance of applying the predictive model to clinical practice.

Most studies based on authoritative databases use standardized and open-access tumor information that is updated regularly, but the question of whether a model using public databases could be used locally should be considered. In addition, some public databases that were established earlier are problematic because clinical practices change over time, and the use of historical data that are too old or a data collection time period that is too long to develop the model will result in the loss of clinical significance [53]. Therefore, researchers should consider focusing more on data management to improve the speed of building models and consider establishing online real-time prediction models. A small number of studies are based on local hospital registration data, but private data require informed consent and ethics committee approval before sharing as well as proper processing (such as anonymity completely). Therefore, the use of private data prevents other scholars from verifying the results of the model and comparing different models.

The number of samples included in this study is uneven. The minimum sample size is 200 patients, and 7 model samples include less than 1000 patients. ML algorithms are often applied to the processing of multidimensional data, and the default application condition is large sample data [54, 55]. The use of too little data in the training model will often lead to overfitting of the model and reduce the generalization ability. In addition, medical data typically contain a large amount of data, outliers, noise redundancy, imbalance, deletion and irrelevant variables [56]. The original dataset will thus cause poor performance of the subsequent prediction model and will become a bottleneck in the process of data mining. Therefore, the process of data preprocessing, including data reduction, data cleaning, data transformation and data integration, is crucial [57] and typically comprises 70~80% of the workload of data mining [58]. However, many of the studies included in this systematic review did not take these key steps. High-quality models depend on high-quality data. In future studies, researchers should not only select appropriate algorithms and perform performance comparisons but also focus on exploring methods for data cleaning and pretreatment and improving the quality and quantity of the modeling data.

Initially, researchers used traditional ML for model construction and then gradually combined and optimized multiple learning models with weak performance to produce ensemble learning algorithms, which have high prediction accuracy and strong generalization ability [59, 60]. However, the above two algorithms are shallow learning algorithms. Although these algorithms play a role, they are often unable to effectively complete tasks such as high-dimensional data processing and large computations when faced with massive data. Therefore, driven by the background of big data cloud computing, deep learning algorithms have been proposed and have gradually become hotspots in breast cancer prediction research. These algorithms are better able to analyze data and model the complex relationship between prognostic variables. The algorithms include factors that depend on time as well as those that interact with other factors associated with prognosis in a nonlinear manner.

In complex modeling problems, there is generally no single algorithm that fits all problems. Different techniques can be combined to produce the best performance, so researchers must compare different ML algorithms or ML algorithms with traditional modeling algorithms. The most commonly used algorithms in this systematic review are ANN, DT, SVM, LR and ensemble learning. Among them, the performance of ANN and DT is better. However, overall, compared with LR/Cox regression model, the performance of the ML algorithm does not necessarily improve, similar to the results of previous studies [61–63].

Model validation is divided into internal and external validation, and internal validation is performed using the dataset randomly obtained from the original dataset, which can be completed by dividing the sample validation. In this study, most of the included studies used cross-validation and random splitting for internal validation, which makes it difficult to avoid overfitting, thus limiting the accuracy of the validation results [64]. External validation requires the development of the queue based on the independence and validation of samples, which is the gold standard of model performance [65, 66]. We found that only 1 study performed external verification of the model. The lack of external validation in multicenter studies with large samples prevents one from determining whether a model is applicable in different scenarios, which can prevent the use of the model, as well as its stability and universality. Thus, data extrapolation should be performed with caution. The lack of practical application of the model in clinical practice may affect the ability of clinicians to make treatment decisions and estimate prognosis. Calibration compares the observed probability and predicted probability of the occurrence of results, which is the key to model development [67]. Only 1 study performed model calibration, and the actual availability of uncalibrated models is limited [68]. Therefore, it is recommended that researchers consider this step and report modeling information in detail.

Compared with the traditional statistical model, the ML algorithm has the black box property. The interpretation and understanding of the model is a key problem [69]. Researchers have difficulties in knowing what happened in the process of prediction and the resulting process, i.e., which variables had the greatest influence on survival and which subgroups of patients showed similar results. Answering these questions can help doctors choose the appropriate treatment and can also eliminate the non-important factors of breast cancer to reduce the time and cost of data collection and treatment. However, whether this problem exists in most models, especially deep learning models, is unknown [70, 71]. The development of this problem and the complex function of internal work are not easy to explain, leading to inappropriate evaluation and feedback to improve the output. In contrast, DT models have excellent interpretability, but their performance still needs to be further optimized [51, 72]. Therefore, compared with focusing only on prediction performance, further understanding of the underlying dynamics of the algorithm has become a research hotspot and led to an increasing number of studies being performed [69].

Regarding factors influencing breast cancer prognosis, screening appropriate predictors as independent variables is an important step in model construction. In previous studies, predictors mostly included patients’ demographic characteristics, medical history, treatment information, and the clinicopathological characteristics of tumors at different disease stages. In this systematic review, we summarize the most commonly used predictors similar to the results of previous studies [4].

Age, disease stage, grade, tumor size, race, marital status, number of nodes, histology, number of positive nodes and primary site code have been entered into many predictive models as predictors, given that these factors represent key risk factors for onset and survival in breast cancer. These variables were also used in studies on decision-making analysis in relation to breast cancer [73–75]. In the future, the possible mechanisms underlying the occurrence and development of breast cancer could be further studied from these perspectives, which also suggests that more suitable predictors for clinical practice can be identified. The ML predictive models applied in this systematic review can be translated into tools for clinical treatment decision-making. Visualization of some of the outcomes will be implemented in the research database and used by the clinicians at the hospital to analyze the survival of breast cancer patients.

With the development of molecular biology, some molecular indicators, such as gene expression and mutation, have also become predictors. Compared with a single data-driven prediction model, in recent years, researchers have incorporated multiple types of data into prediction. The rapid increase in the number of features from different data sources and the use of heterogeneous features have led to great challenges in survival prediction. With the deepening of research on breast cancer, many new variables that are significantly related to breast cancer prognosis have been gradually discovered [14, 76, 77], such as the level of anxiety and depression. Thus, the above factors should be taken into account in the prediction. This notion illustrates the true complexity of breast cancer as a disease, highlights the importance of the mechanisms involved, and highlights some of the confusion among researchers in selecting the most appropriate prediction model.

The number of candidate predictors and the correlation between them will affect model performance. Therefore, feature selection becomes particularly important. Feature selection identifies the most important variables in the dataset while maintaining classification quality. A reduction in the number of predictors and the burden of data collection can reduce the fitting and complexity of the model and help researchers interpret and understand the model. However, many studies did not report the feature selection process, which may be related to the ability of some classification models to deal with high-dimensional datasets (e.g., RF, SVM, DT), or the features included in the models may have been selected based on prior research or clinical importance [14, 78].

No quality assessment criteria have been established specifically for systematic reviews of ML research. Existing guidelines, such as CHARMS [21] and TRIPOD [79], do not consider the characteristics and related biases of ML models. There have been studies using improved quality assessment criteria to adapt to ML system evaluation [62, 80, 81], but they have not been widely accepted. Therefore, with the increasing application of ML in prediction and other fields, it is recommended that guidelines be developed for reporting and evaluating ML prediction model research in the medical field and to serve as a standard for publication to improve the quality of related papers.

Limitations

This study has some limitations. First, only English studies were included, so publication bias may be present. Second, the excessive differences of the included studies limit the comparison between studies and prohibits the use of meta-analysis [82, 83]. Finally, most of the included studies did not report the key steps in model development and validation. In addition, the information on predictive performance (such as true positive, false positive, true negative, and false negative in the confusion matrix) was insufficient, and most of the studies only described a single dimension of predictive performance. Therefore, it is recommended that comprehensive methodological information, such as missing value processing, outlier value processing, class imbalance processing, hyperparameter tuning, feature selection variable importance ranking processing, model evaluation and validation, be reported in detail along with the model performance, including detailed information on the suitability and acceptability of classification, discrimination and calibration measures.

Conclusion

ML has become a new methodology for breast cancer survival prediction, and there is still much room for improvement and potential for further model construction. The existing prediction models still face limitations related to a lack of data preprocessing steps, the excessive differences of sample feature selection, and issues related to validation and promotion. The model performance still needs to be further optimized, and other barriers should be addressed. Researchers and medical workers should connect with reality, choose a model carefully, use the model in clinical practice after verification, and use rigorous design and validation methods with a large sample of high-quality research data on the basis of previous findings. The applicability and limitation of these models should be evaluated strictly to improve the degree of accuracy for breast cancer survival prediction.