Skip to main content

Machine learning-based techniques to improve lung transplantation outcomes and complications: a systematic review

Abstract

Background

Machine learning has been used to develop predictive models to support clinicians in making better and more reliable decisions. The high volume of collected data in the lung transplant process makes it possible to extract hidden patterns by applying machine learning methods. Our study aims to investigate the application of machine learning methods in lung transplantation.

Method

A systematic search was conducted in five electronic databases from January 2000 to June 2022. Then, the title, abstracts, and full text of extracted articles were screened based on the PRISMA checklist. Then, eligible articles were selected according to inclusion criteria. The information regarding developed models was extracted from reviewed articles using a data extraction sheet.

Results

Searches yielded 414 citations. Of them, 136 studies were excluded after the title and abstract screening. Finally, 16 articles were determined as eligible studies that met our inclusion criteria. The objectives of eligible articles are classified into eight main categories. The applied machine learning methods include the Support vector machine (SVM) (n = 5, 31.25%) technique, logistic regression (n = 4, 25%), Random Forests (RF) (n = 4, 25%), Bayesian network (BN) (n = 3, 18.75%), linear regression (LR) (n = 3, 18.75%), Decision Tree (DT) (n = 3, 18.75%), neural networks (n = 3, 18.75%), Markov Model (n = 1, 6.25%), KNN (n = 1, 6.25%), K-means (n = 1, 6.25%), Gradient Boosting trees (XGBoost) (n = 1, 6.25%), and Convolutional Neural Network (CNN) (n = 1, 6.25%). Most studies (n = 11) employed more than one machine learning technique or combination of different techniques to make their models. The data obtained from pulmonary function tests were the most used as input variables in predictive model development. Most studies (n = 10) used only post-transplant patient information to develop their models. Also, UNOS was recognized as the most desirable data source in the reviewed articles. In most cases, clinicians succeeded to predict acute diseases incidence after lung transplantation (n = 4) or estimate survival rate (n = 4) by developing machine learning models.

Conclusion

The outcomes of these developed prediction models could aid clinicians to make better and more reliable decisions by extracting new knowledge from the huge volume of lung transplantation data.

Peer Review reports

Introduction

The last decade has seen a gradual but significant increase in organ transplants, although lung transplants have a lower rate than other organ transplants [1]. Lung transplantation (LTx) is a well-established treatment for a wide variety of end-stage lung diseases [2, 3]. By 2019, more than 4,500 lung transplants were executed at 260 centers worldwide, according to the International Heart and Lung Transplant Association (ISHLT) [4]. Despite all advances in medicine, LTx encounters some difficulties like lung allograft dysfunction, organ rejection, side effects of immunosuppressive therapy, or complex infection yet [5]. Nowadays, predicting transplant complications, determining risk factors, and estimating the success rate of transplants are the main concerns of physicians in this field with the increase in transplant cases. Developing prediction models could aid clinicians in organ allocation and estimating post-transplantation outcomes and complications [6]. Though traditional statistical methods have not been able to meet their needs, some predictive models were developed to estimate the post-transplant survival rate or side effects with the hope of increasing the success of transplantation using Artificial Intelligence (AI) or machine learning (ML) techniques.

It seems that the development of data-driven ML techniques can support clinicians in making more informed decisions by generating new insights into disease in medicine [7]. Machine learning (ML) techniques are a set of methods for analyzing a large amount of data to reveal hidden patterns in data sets or explain the relationship between various variables [8]. Machine learning methods in medicine have been applied in a wide range of areas such as cancer problem-solving, medicinal chemistry, brain and neurology, medical imaging, and data analysis of wearable sensors for symptom monitoring [9]. ML methods can deal with large and complex medical data and analyze them easily to find new ways for accurate diagnosis and treating patients [10, 11]. Utilizing ML models in other organ transplants could estimate the risk of acute rejection, survival rate, risk factors, and prevalent comorbidities after transplantation [12]. In addition, they can determine the most appropriate organ recipient and those at high risk of post-transplant mortality by building ML models based on various parameters [13].

Many efforts have been made to develop predictive models in LTx using machine learning techniques [13,14,15]. No study has been published to investigate the applied methods in this domain. Our study aims to systematically review all published evidence on the utilization of ML techniques as one of the main approaches of artificial intelligence in lung transplantation. In addition to providing a comprehensive overview of the most widely used machine learning methods in LTx, our additional objectives include identifying the main challenges and concerns that machine learning methods are being built to deal with them.

Method

Research questions

This study was conducted to answer the following research questions:

R-Q1- Which machine learning techniques are used in the lung transplantation domain?

R-Q2- For which specific problems and objectives in lung transplantation has machine learning been used?

R-Q3- In which stages of lung transplantation have machine learning methods been applied?

R-Q4- What data sources or databases have been used in lung transplantation to develop machine-learning models?

RQ-5- Which features are used by the machine learning models reviewed studies?

RQ-6-Which evaluation techniques have been employed in developed models?

Search strategy

A systematic search was conducted in six electronic databases, Medline (PubMed), Scopus, Web of Science (WOS), PsycINFO, IEEE, and Cochrane databases using pre-specified search strategies based on keywords. Database searching was performed in the period from January 2000 to June 2022. Reference lists of retrieved articles and review articles in the field were also searched to identify eligible studies that met inclusion and exclusion criteria. This systematic review was conducted using the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) checklist [16].

Inclusion and exclusion criteria

Our research questions were outlined based on PCC (Population, Concept, Context) to conduct qualitative review studies [17]. Population refers to lung transplant recipients, candidates for lung transplantation, or those on the waiting list. The concept referred to lung transplantation and all related complications, outcomes, side effects, and affective factors. Context referred to any machine learning techniques applied in LTx.

We included all full-text articles focusing on the utilization of ML techniques in lung transplantation. The inclusion criteria for this review were as follows: (1) original research study, (2) Studies to be included if it has provided sufficient information on the machine learning algorithms used for the analysis, (3) Article is included if it evaluated the applied ML techniques, (4) Topics related to lung transplantation, (4) All types of lung transplant recipients (single or double), candidates, or patients in waiting list.

Non-peer-reviewed articles, all types of review articles, meta-analyses, letters to the editor, commentaries, abstracts, editorials, patents, perspectives, or studies with non-human species were excluded. Studies were excluded too if they (1) were not full-length publications, (2) The protocol or methods papers, (3) Not English papers, and (4) Machine Learning methods were not used at all.

Data extraction process

Two researchers screened the title and abstracts of extracted articles, independently. The screening process was done using the PRISMA checklist. Then, two researchers (MG and RS) read the full texts. The disagreement was resolved by the supervision of the other researcher (HA).

The details of the methodology and outcomes of reviewed articles were noted on a data extraction sheet. The data extraction process was done by two reviewers. Data elements extracted included title, publication year, country, population, feature selection technique, input variables, applied ML method, training and validation size, validation techniques, and model performance indicators.

Risk of bias

The risk of bias in included articles was evaluated by two independent reviewers using the Cochrane collaboration risk of bias tool suggested by Narukab [14] for ML-related articles. The methodological quality of the articles was assessed based on these domains: (1) Data collection, (2) Study Response, (3) Outcome Measurement, (4) and Statistical Analysis and Reporting. The high risk, not clear, and low risk of bias was allocated to each study.

Quality assessment

Due to the heterogeneity of reviewed studies, the quantitative analysis was inappropriate. Hence, the quality of reviewed articles was evaluated by a quality assessment table for machine learning studies suggested by Qiao [18]. According to Qiao’s study, articles in the machine learning field are reviewed in nine areas in terms of quality. These categories include limits in the current non-machine learning approach, valid methods for over-fitting, predictors for an explanation, hyperparameters, using external data validation, feature engineering methods, applied platforms, stability of results, and suggested clinical use.

Results

Searches yielded 414 citations. Of them, 185 articles remained after duplication removal. From 185 retrieved articles, 136 studies were excluded after the title and abstract screening because their topics were inapplicable to our subject. Next, the full text of 49 articles was reviewed according to inclusion criteria. Finally, 16 articles were determined as eligible studies that met our inclusion criteria. The screening process and the reasons for deleting articles following the PRISMA report checklist are described in Fig. 1.

Fig. 1
figure 1

The flow diagram of article screening according to the PRISMA checklist

Of 16 studies, 12 (75%) of the articles were published after 2015 [13, 19,20,21,22,23,24,25,26,27,28]. In terms of country, ten studies were conducted in the USA [13, 20, 22, 25, 28,29,30,31,32,33], and one study each in Belgium [21], China [27], Iran [19], Italy [23], Spain [24], and UK [26]. The summaries of applied techniques and characteristics of articles are described in Table 1.

Table 1 Summaries of reviewed articles

machine learning methods applied in reviewed articles

Among 16 papers, the Support vector machine (SVM) technique (n = 5, 31.25%) is the most favorable method utilized by the authors [21, 22, 26, 31, 33]. After that, logistic regression (n = 4, 25%) [24, 25, 30, 33], and random forests (RF) (n = 4, 25%) [13, 23, 27, 28] have been the most used techniques to develop data-driven models concerning lung transplantation. Bayesian network (BN) (n = 3, 18.75%) [19, 29, 32], linear regression (LR) (n = 3, 18.75%) [13, 26, 28], Decision Tree (DT) (n = 3, 18.75%) [30, 31], neural networks (n = 3, 18.75%) [24, 30, 31], Markov Model (n = 1, 6.25%) [29], KNN (n = 1, 6.25%) [31], K-means (n = 1, 6.25%) [31], Gradient Boosting trees (XGBoost) (n = 1, 6.25%) [20], and Convolutional Neural Network (CNN) (n = 1, 6.25%) [25] were other strategies used to develop machine learning models in studies. Only one article employed a deep learning method for image processing using CT features [25].

A brief description of applied machine learning techniques is represented in Table 2. Most studies (n = 11) employed more than one machine learning technique or combination of different techniques to make their models [13, 20, 22,23,24,25, 28,29,30,31, 33].

Table 2  A summary of the machine learning methods employed in LTx

In all articles, ML methods can be divided into two broad categories, supervised and unsupervised learning techniques. Though in the machine learning field, there is another category for these techniques called transfer learning which was not employed in the reviewed studies. Most of the studies took only supervised machine learning techniques (Fig. 2). Only two studies have employed the combination of supervised and unsupervised learning techniques in reviewed articles.

Fig. 2
figure 2

Proportion of various ML methods used in the literature

specific problems and objectives in lung transplantation

All studies applied ML methods with different approaches. The objectives of eligible articles are classified into eight main categories. All reviewed articles tried to solve some problems using machine learning techniques, these objectives and their frequency are described in Table 3. In most cases, developing ML models can help clinicians to predict acute diseases incidence after lung transplantation (n = 4) or estimate survival rate (n = 4).

Table 3 General characteristics of reviewed articles

The distribution of applied ML techniques in terms of their objectives is represented in Fig. 3.

Fig. 3
figure 3

Distribution of ML methods in terms of their objectives

Predicting the acute disease events after transplantation

The occurrence of some diseases after lung transplantation is inevitable. Evidence showed that more than 50% of lung recipients experienced bronchiolitis obliterans syndrome (BOS) development after transplantation surgery by five years [34]. Of 16 studies, four articles were devoted to improving the predicting BOS incidence after LTx. Troiani [29] and Hosseini-Baharanchi [19] applied the Bayesian network model to predict BOS incidence after LTx. Troiani [29] used spirometry variables and symptom covariates to predict BOS to compare different models. Their results showed that the Bayesian approach was preferable to classical approaches in BOS classification in lung transplant recipients. Similarly, Hosseini-Baharanchi [19] used demographic data and some clinical variables to predict BOS incidence after LTx. They found that CMV infection was a good predictor of BOS incidence in LTx patients.

Barbosa et al. [21, 33] developed classification models to predict BOS development using the SVM technique and logistic regression based on CT features in two studies. Their results showed that the combination of CT metrics with PFT (Pulmonary function tests) as predictors could enhance model diagnostic accuracy for all transplant types.

Predicting the survival rate of post-transplantation

However, the survival rate of lung transplants compared to other organ transplants is low. But according to the International Society of Heart and Lung Transplantation (ISHLT) Report, the existing five-year survival rate after lung transplantation is approximately 55% [35]. Four articles are devoted to predicting and estimating the survival rate after LTx.

The results of the Oztekin.A et al. [30] the study showed using data-mining methodology could support clinicians to select more related variables which were effective in predicting graft survival. According to their study, the integration of neural network models, decision trees, and logistic regression with the Cox hazard model was able to show results with satisfactory prediction accuracy compared to the traditional methods that were used before.

Oztekin. A [32] and his collagenous in another study in 2011 succeed to develop a decision support tool using decision trees and neural networks. By designing this decision-making tool based on data mining models, the doctor in the case of organ donation can quickly decide which patient is the most suitable potential recipient for donor organ allocation.

Mark.E et al. [13] developed machine learning-based models to estimate 5-year survival rates for patients using linear regression and Random Forest. According to their models, patients are predicted to have a higher predicted survival accepting an increased risk for disease transmission (IRD) organ offer compared to waiting for a non–IRD organ with average wait times.

Shaish et al. [25] created his model using a deep learning method (CNN) to classify the patterns of disease progression for Usual interstitial pneumonia (UIP) patients and determine the severity of the disease with survival rate after LTx. Their results showed that CNN-predicted UIP was associated with an increased risk of death or lung transplantation during cross-validation.

Predicting recipient-donor matching

Match-to-recipient (D/R) in lung transplantation (LTx) is usually determined based on blood group and predicted total lung capacity (pTLC), as well as height and age [36, 37]. Predicting the recipient-donor matching and predicting the most important factors could be beneficial for clinicians in selecting the most suitable recipient. Dueñas-Jurado [24] and Zafar. F [28] developed prediction models based on the characteristics of donor recipients and past experiences with lung donors and recipients to improve donor-recipient matching in lung transplantations.

Dueñas-Jurado et al. [24] created a model in the lung allocation system for matching lung transplant donor recipients using neural networks. The predictors used to predict the probability of survival rate and recipient-donor matching included lower pre-transplant carbon dioxide (PCO2) pressure, higher pre-transplant and post-transplant functional vital capacity (FVC), lower donor mechanical ventilation, and shorter ischemia time. The proposed model represented in this study was a powerful tool for donor-recipient matching that showed higher accuracy than classical statistical methods.

Similarly, Zafar. F [28] developed a comprehensive model to guide recipient-donor matching using random forest and cox regression using clinical and demographic data of recipients and donors. They developed an online Lung Transplantation Advanced Prediction Tool (LAPT) in the form of a simple calculator. Employing this tool, users can enter recipient and donor information to calculate predicted 1-, 5-, and 10-year survival, risk stratification, and associated survival and half-life predictions. Top selected common features that are effective in predicting quality of life and identified through this study included simultaneous lung, type of transplant, recipient CMV results at transplantation, recipient CMV results at transplantation, any drug-treated infection, chronic steroid at transplant, recipient age, prior cardiac surgery at transplant, and infection requiring IV drug therapy.

Determining the relation between pulmonary function tests and LTx outcomes

Predicting Total Lung Capacity (pTLC) has a significant role in LTx outcomes. Hence, Pande. A et al. [20] explored the relationship between Forced expiratory volume in the first second (FEV1) and age of lung recipients with LTx status using the novel multivariate tree boosting method on longitudinal data of spirometry tests. Their investigation using FEV1 longitudinal data and application of the feature selection method revealed that double-lung recipients not only have a higher FEV1 but also have a slower decline in lung capacity than single-lung recipients. They succeeded to apply a novel multivariate tree-boosting method for fitting a semi-nonparametric model.

In another study, they investigated the role of cardiopulmonary exercise testing (CPET) parameters on symptoms of lung recipients after transplantation using a random forest tree [23]. Developing predictive models revealed that muscle pain at peak exercise was strongly associated with altered basal and exercise-induced metabolic pathways. The onset of dyspnea was associated with the intensity of the ventilatory response to meet the metabolic demands of increased workloads.

Predicting the most important reasons for transplant rejection

Rejection is a major complication that remains an important problem after lung transplantation. Despite advances in immunosuppressive therapy and immunosuppressive drugs used, more than one-third of lung transplant recipients are treated for acute rejection in the first year after transplantation [38]. Some of the most important reasons for transplant rejection in patients are infection and primary graft dysfunction after lung transplantation. In this regard, Su. J et al. [27] utilized a random forest model to survey the association between airway infection and rejection in lung transplant recipients (LTRs). Developing models revealed the role of airway microbiota, especially together with PCT and T-lymphocyte levels in differentiating between clinically stable recipients and those with infection and acute rejection.

Stefanuto.P et al. [26] made prediction models using SVM techniques and multivariant analysis to discover the relation of pulmonary volatile organic compound (VOC) spectrum with primary graft dysfunction in lung transplant recipients. The authors succeeded to develop a model which identified patients with grade 3 primary graft dysfunction (PGD) with an AUROC of 0.90 and a positive predictive value of 0.88. This indefinable molecular approach represents a new molecular strategy for detecting and monitoring allograft injury.

Predict the risk factors

Determining the risk factors for each patient before LTx surgery can support clinicians decide to what extent each transplantation is beneficial for each patient. Hence, Delen.A et al. [31] tried to determine predictor variables and risk factors affecting survival time through various machine learning methods (SVM, ANN, MLP, RBF, DT, and K-means for clustering the results). This study identified a group of risk factors and a comprehensive list of predictors in graft survival. Some factors such as age, gender of the recipient, and his medical condition at admission time were discussed in previous studies. While others have been neglected in previous studies such as the recipient’s length of stay after transplantation, and the interaction of gender and ethnicity between the recipient and the donor.

Predicting quality of life after transplantation

Lung transplantation usually has significant effects on a patient’s health-related quality of life (HRQL). Patients’ satisfaction with the quality of life can affect their physical, emotional health, social, and sexual functioning. Additionally, it aids them to cope with their lives easily [39,40,41]. One of the main objectives of LTx is to enhance the quality of life in end-stage lung disease in addition to survival rate increment [39]. Thus, Oztekin.A et al. [22] (2018) compared various models like SVM, KNN, and neural networks to predict the quality of life after lung transplantation. The evaluation of three developed models proved that the hybrid GA-SVM model has high performance in comparison to the other two models to predict the quality of life after lung transplantation.

Different stages of lung transplantation using machine learning methods

The authors utilized the data from donors, candidates, and transplant patients to develop their models. Six studies developed their models based on both pre-and post-transplant patient information [13, 21, 22, 25, 27, 28], and ten studies used only post-transplant patient information to develop their models [19, 20, 23, 24, 26, 29,30,31,32,33]. Two studies used related data from donor and transplant patients [24, 28]. The distribution of articles based on the transplantation phase is shown in Fig. 4.

Fig. 4
figure 4

The distribution of articles based on the transplantation phase

Features used in the ML algorithm

Machine learning or data mining algorithms use a series of features as input or predictive factors to build models or classify output variables. These features are selected according to the objective of the researcher to produce the output. Table 4 shows the features used to develop each model and the final goal of creating the model.

Most of the articles (n = 12) employed clinical data including demographic data, laboratory data, pulmonary function test results, and follow-up data as input variables, while three articles utilized CT features and images in combination with pulmonary function test results as input variables to create a prediction model. Only one study used CT images to devise a new prediction model [21, 25, 33]. As we can see, the data obtained from pulmonary function tests were the most used in predictive model development.

Table 4 Independent variables or features used in machine learning algorithms as input variables

Despite the high number of features in the dataset, only nine papers applied the feature selection method [13, 20, 22, 23, 26,27,28, 31]. Feature selection helps researchers select the best set of features to build valuable models of the subjects under study. The applied methods for feature selection include the k-means algorithm (n = 1) [31], permutation variable importance (VIMP) (n = 1) [20], genetic algorithm (n = 1) [22], random forest algorithm (n = 3) [13, 23, 27], LASSO Cox regression (n = 1) [28], and SVM technique (n = 1) [26].

Six of 16 articles developed their models using data sets including more than 1500 samples with more than 30 features [13, 22, 28, 30, 31, 42]. Four studies developed their model using datasets between 1,000 and 100 patients [20, 24, 25, 33], and six studies with less than 100 patients [19, 21, 23, 26, 27, 29]. The sample size varied from 16 to 310,773 individuals.

The data sources

A source of data or a set of data is needed to develop a model. Data sets are usually large databases of data that are collected and organized for a specific purpose. In terms of data sources, six studies employed the open-source United Network for Organ Sharing (UNOS) dataset to create their models [13, 22, 28, 30,31,32]. The UNOS dataset is freely available to researchers and has been used in many important linkage studies. UNOS gathers all transplant-related data on every U.S. organ donor, transplant candidate, recipient, and consequence [43]. Other studies have used local databases to build their models [19,20,21, 23,24,25,26,27, 29, 33]. The data sources with their frequency is described in Table 5.

Table 5 Database/Data sources used as data sources in developed models

The performance of developed models and evaluation methods

In machine learning, we have to evaluate the stability of developed models to estimate the generalization accuracy of a model’s unseen data. In developing a machine learning model, it is important that the model created is not over-fitting or under-fitting. While underfitting is usually the result of not training the model with enough data, overfitting can have a variety of causes. The ultimate goal of machine learning is to develop a model that performs well with both training data and new data used for predictions. There are two common approaches to evaluating models including creating a holdout dataset or performing cross-validation. Hold-out refers to the strategy in which we divide the data set into a train and a test set. Cross-validation is a technique that involves dividing the original observation data set into a training set that is used to train the model and an independent set that is used to evaluate the analysis. The most common cross-validation method is k-fold cross-validation. In k-fold cross-validation, we can split the input data into k subsets of data. Of 16 articles, 14 articles employed the cross-validation method to detect overfitting and validate the model efficiency. Eight studies utilized a train-test split to evaluate the machine learning algorithm.

To evaluate the performance of developed models, various metrics were applied including calculating accuracy, sensitivity, F-measure, Root Mean Square Error (RMSE), R-squared (R2), correlation matrix, chi-square, or AUC (Area Under the Receiver Operating Characteristics). The most common metric used in studies to evaluate the model alone or to compare it with other developed models was accuracy (n = 7). Next, AUC was applied in five studies. The frequency of applied metrics is shown in Fig. 5. The explanation of applied metrics is described in Table 6.

Fig. 5
figure 5

The distribution of applied metrics in reviewed articles

Among seven studies that applied the accuracy metric, five studies this metric in combination with sensitivity and specificity. In one study, the F-measure score was utilized to evaluate the models in combination with accuracy, sensitivity, and specificity.

Table 6 The most common metric used in studies

The quality assessment of reviewed articles

The reviewed articles were evaluated in nine categories suggested by Qiao [18]. The results of the evaluation of articles based on these nine axes are shown in Table 7.

Table 7 The results of the quality assessment of reviewed articles

The summative score of all articles was above five. The results of the risk of bias assessment in the reviewed articles are shown in Fig 6.

Fig. 6
figure 6

Risk of bias summary regarding each risk of bias item for each included study

Discussion

Our systematic review investigated utilizing machine learning in a lung transplant domain. According to the PRISMA checklist, sixteen papers were recognized as eligible articles to identify the usage of machine learning in lung transplantation. Evidence showed that the development of machine learning models in organ transplantation had surprising results in improving transplant success and predicting the likelihood of transplant rejection [15, 44, 45]. Although machine learning techniques can support clinicians predict transplant complications, limited studies have been conducted in lung transplant domains [46].

The majority of the reviewed articles were dedicated to the use of ML models in predicting survival and comorbidities [13, 19, 21, 25, 29, 30, 32, 33]. However, several studies used ML techniques to predict mortality after receiving transplanted organs [44, 45], but no study was done in LTx to predict mortality. Despite the importance of predicting the degree of donor and recipient matching to increase the success of transplant surgery, only two articles were devoted to this topic [24, 28]. Comprehending the benefits of analyzing the main results of LTx with the aid of the machine learning method can make clinicians interested in using this new method based on artificial intelligence.

ML development works best with a high volume of data samples and a rich set of features. Thus, the majority of articles developed their models using UNOS large datasets [47]. Valuable studies have been conducted using UNOS datasets because all of these kinds of data are available for all researchers free of charge [48]. Thus, designing the integrated large database to record all LTx-related data could aid researchers to conduct big data analytics in LTx.

As stated in the results section, most applied techniques were supervised learning methods like SVM, DT, RF, and regression. Although deep learning methods play an essential role in medicine due to their ability to analyze complex data [49], only one study was devoted to deep learning techniques to analyze medical images.

Overfitting is one of the main challenges in supervised ML methods which prevents generalizability [50]. Due to a large amount of LTx data, the authors are faced with high-dimensional data for developing their methods. Usually, all data features are not useful. Hence, feature selection methods to select the best set of features could be effective to design more best-fitting models in medical sciences [51, 52]. Despite all of these benefits, only nine papers applied these kinds of methods. Different metrics have been employed to evaluate the developed ML models. Because the same metrics have not been used to compare the performance of the developed models, it is not possible to compare these techniques. Although the evaluation criteria were mentioned in some studies, the evaluation results are not stated clearly in the articles.

Implementing clinical decision support tools (CDSS) based on developed ML models could support healthcare providers to deliver optimal care to patients [53]. They can analyze a large amount of data in the shortest time and suggest the best treatment options to the physicians [54]. Despite these advantages, no study has designed a decision support system in this field.

This review also faced some limitations. Although search strategies were done in different databases, all machine learning models in this field might not be identified through a literature review. Researchers may have reported only high-performing models. So, publication bias could exist in this review. The author of the studies did not consider all available variables from data sources to develop their models. Considering new variables with the same goals and same data sources might be generated new models with different accuracy. Another limitation of this study is that we considered a wide classification domain of machine learning algorithms in lung transplantation with diverse data sources. Therefore, we cannot compare them in terms of performance. We did not consider any subclassification of any of the considered algorithms or data sources in this study. Ultimately, the development and application of deep learning should be considered the main subject of further study.

Conclusion

This review showed that applying ML methods could target clinical problems and complications in lung transplantation as one of the complex subjects in medicine. The outcomes of these developed prediction models could aid clinicians to make better and more reliable decisions by extracting new knowledge from the huge volume of data. Deep learning method utilization in lung transplant data analysis could be the main research gap in this field, which can be the main topic of future studies.

Availability of data and materials

The datasets generated and/or analyzed during the current study are not publicly available but are available from the corresponding author at reasonable request.

Abbreviations

SVM:

Support vector machine

RF:

Random Forests

BN:

Bayesian network

DT:

Decision Tree

KNN:

K-nearest neighbors

XGBoost:

Gradient Boosting trees

CNN:

Convolutional Neural Network

References

  1. Rampolla R. Lung transplantation: an overview of candidacy and outcomes. Ochsner J. 2014;14(4):641–8.

    Google Scholar 

  2. Meyer KC. Recent advances in lung transplantation. F1000Res 2018, 7.

  3. Studer SM, Levy RD, McNeil K, Orens JB. Lung transplant outcomes: a review of survival, graft function, physiology, health-related quality of life and cost-effectiveness. Eur Respir J. 2004;24(4):674–85.

    Article  CAS  Google Scholar 

  4. Leard LE, Holm AM, Valapour M, Glanville AR, Attawar S, Aversa M, Campos SV, Christon LM, Cypel M, Dellgren G, et al. Consensus document for the selection of lung transplant candidates: an update from the International Society for Heart and Lung Transplantation. J Heart Lung Transplant. 2021;40(11):1349–79.

    Article  Google Scholar 

  5. Bos S, Vos R, Van Raemdonck DE, Verleden GM. Survival in adult lung transplantation: where are we in 2020? Curr Opin Organ Transplant. 2020;25(3):268–73.

    Article  Google Scholar 

  6. Levvey B, Keshavjee S, Cypel M, Robinson A, Erasmus M, Glanville A, Hopkins P, Musk M, Hertz M, McCurry K. Influence of lung donor agonal and warm ischemic times on early mortality: analyses from the ISHLT DCD Lung Transplant Registry. J Heart Lung Transplantation. 2019;38(1):26–34.

    Article  Google Scholar 

  7. Medved D, Ohlsson M, Höglund P, Andersson B, Nugues P, Nilsson J. Improving prediction of heart transplantation outcome using deep learning techniques. Sci Rep. 2018;8(1):3613.

    Article  Google Scholar 

  8. Sidey-Gibbons JAM, Sidey-Gibbons CJ. Machine learning in medicine: a practical introduction. BMC Med Res Methodol. 2019;19(1):64.

    Article  Google Scholar 

  9. Shehab M, Abualigah L, Shambour Q, Abu-Hashem MA, Shambour MKY, Alsalibi AI, Gandomi AH. Machine learning in medical applications: a review of state-of-the-art methods. Comput Biol Med. 2022;145:105458.

    Article  Google Scholar 

  10. Handelman GS, Kok HK, Chandra RV, Razavi AH, Lee MJ, Asadi H. eDoctor: machine learning and the future of medicine. J Intern Med. 2018;284(6):603–19.

    Article  CAS  Google Scholar 

  11. Kumar CJ, Das PR. The diagnosis of ASD using multiple machine learning techniques. International Journal of Developmental Disabilities 2021:1–11.

  12. Killian MO, Payrovnaziri SN, Gupta D, Desai D, He Z. Machine learning–based prediction of health outcomes in pediatric organ transplantation recipients. JAMIA. 2021;4(1):ooab008.

    Google Scholar 

  13. Mark E, Goldsman D, Keskinocak P, Sokol J. Using machine learning to estimate survival curves for patients receiving an increased risk for disease transmission heart, liver, or lung versus waiting for a standard organ. Transpl Infect Dis. 2019;21(6):e13181.

    Article  Google Scholar 

  14. Naruka V, Arjomandi Rad A, Subbiah Ponniah H, Francis J, Vardanyan R, Tasoudis P, Magouliotis DE, Lazopoulos GL, Salmasi MY, Athanasiou T. Machine learning and artificial intelligence in cardiac transplantation: A systematic review. Artificial Organs, n/a(n/a).

  15. Ferrarese A, Sartori G, Orrù G, Frigo AC, Pelizzaro F, Burra P, Senzolo M. Machine learning in liver transplantation: a tool for some unsolved questions? Transpl Int. 2021;34(3):398–411.

    Article  Google Scholar 

  16. Moher D, Liberati A, Tetzlaff J, Altman DG. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. BMJ. 2009;339:b2535.

    Article  Google Scholar 

  17. Institute JB. Joanna Briggs Institute reviewers’ manual: 2014 edition. Australia: The Joanna Briggs Institute 2014:88–91.

  18. Qiao N. A systematic review on machine learning in sellar region diseases: quality and reporting items. Endocr Connect. 2019;8(7):952–60.

    Article  Google Scholar 

  19. Hosseini-Baharanchi FS, Hajizadeh E, Baghestani AR, Najafizadeh K, Shafaghi S. Bronchiolitis obliterans syndrome and death in iranian lung transplant recipients: a bayesian competing risks analysis. Tanaffos. 2016;15(3):141–6.

    Google Scholar 

  20. Pande A, Li L, Rajeswaran J, Ehrlinger J, Kogalur UB, Blackstone EH, Ishwaran H. Boosted Multivariate Trees for Longitudinal Data. Mach Learn. 2017;106(2):277–305.

    Article  Google Scholar 

  21. Barbosa EJM Jr, Lanclus M, Vos W, Van Holsbeke C, De Backer W, De Backer J, Lee J. Machine learning algorithms utilizing quantitative CT features may predict eventual onset of Bronchiolitis Obliterans Syndrome after Lung Transplantation. Acad Radiol. 2018;25(9):1201–12.

    Article  Google Scholar 

  22. Oztekin A, Al-Ebbini L, Sevkli Z, Delen D. A decision analytic approach to predicting quality of life for lung transplant recipients: a hybrid genetic algorithms-based methodology. Eur J Oper Res. 2018;266(2):639–51.

    Article  Google Scholar 

  23. Braccioni F, Bottigliengo D, Ermolao A, Schiavon M, Loy M, Marchi MR, Gregori D, Rea F, Vianello A. Dyspnea, effort and muscle pain during exercise in lung transplant recipients: an analysis of their association with cardiopulmonary function parameters using machine learning. Respir Res. 2020;21(1):267.

    Article  CAS  Google Scholar 

  24. Dueñas-Jurado JM, Gutiérrez PA, Casado-Adam A, Santos-Luna F, Salvatierra-Velázquez A, Cárcel S, Robles-Arista CJC, Hervás-Martínez C. New models for donor-recipient matching in lung transplantations. PLoS ONE. 2021;16(6):e0252148.

    Article  Google Scholar 

  25. Shaish H, Ahmed FS, Lederer D, D’Souza B, Armenta P, Salvatore M, Saqi A, Huang S, Jambawalikar S, Mutasa S. Deep learning of computed Tomography virtual wedge resection for prediction of histologic usual interstitial pneumonitis. Ann Am Thorac Soc. 2021;18(1):51–9.

    Article  Google Scholar 

  26. Stefanuto PH, Romano R, Rees CA, Nasir M, Thakuria L, Simon A, Reed AK, Marczin N, Hill JE. Volatile organic compound profiling to explore primary graft dysfunction after lung transplantation. Sci Rep. 2022;12(1):2053.

    Article  CAS  Google Scholar 

  27. Su J, Li CX, Liu HY, Lian QY, Chen A, You ZX, Li K, Cai YH, Lin YX, Pan JB, et al. The Airway Microbiota Signatures of infection and rejection in lung transplant recipients. Microbiol Spectr. 2022;10(2):e0034421.

    Article  Google Scholar 

  28. Zafar F, Hossain MM, Zhang Y, Dani A, Schecter M, Hayes D Jr, Macaluso M, Towe C, Morales DLS: Lung Transplantation Advanced Prediction Tool: Determining Recipient’s Outcome for a Certain Donor. Transplantation 2022.

  29. Troiani JS, Carlin BP. Comparison of bayesian, classical, and heuristic approaches in identifying acute disease events in lung transplant recipients. Stat Med. 2004;23(5):803–24.

    Article  Google Scholar 

  30. Oztekin A, Delen D, Kong Z. Predicting the graft survival for heart-lung transplantation patients: an integrated data mining methodology. Int J Med Informatics. 2009;78(12):e84–96.

    Article  Google Scholar 

  31. Delen D, Oztekin A, Kong Z. A machine learning-based approach to prognostic analysis of thoracic transplantations. Artif Intell Med. 2010;49(1):33–42.

    Article  Google Scholar 

  32. Oztekin A, Kong ZYJ, Delen D. Development of a structural equation modeling-based decision tree methodology for the analysis of lung transplantations. Decis Support Syst. 2011;51(1):155–66.

    Article  Google Scholar 

  33. Barbosa EM Jr, Simpson S, Lee JC, Tustison N, Gee J, Shou H. Multivariate modeling using quantitative CT metrics may improve accuracy of diagnosis of bronchiolitis obliterans syndrome after lung transplantation. Comput Biol Med. 2017;89:275–81.

    Article  Google Scholar 

  34. Sithamparanathan S, Thirugnanasothy L, Clark S, Dark JH, Fisher AJ, Gould KF, Hasan A, Lordan JL, Meachery G, Parry G, et al. Observational study of lung transplant recipients surviving 20 years. Respir Med. 2016;117:103–8.

    Article  Google Scholar 

  35. Chen-Yoshikawa TF. Ischemia-Reperfusion Injury in Lung Transplantation. Cells. 2021;10(6):1333.

    Article  Google Scholar 

  36. Briceño J, Ciria R, de la Mata M. Donor-recipient matching: myths and realities. J Hepatol. 2013;58(4):811–20.

    Article  Google Scholar 

  37. Demir A, Coosemans W, Decaluwé H, De Leyn P, Nafteux P, Van Veer H, Verleden GM, Van Raemdonck D. Donor-recipient matching in lung transplantation: which variables are important?†. Eur J Cardiothorac Surg. 2015;47(6):974–83.

    Article  Google Scholar 

  38. Parulekar AD, Kao CC. Detection, classification, and management of rejection after lung transplantation. J Thorac Dis. 2019;11(Suppl 14):1732-s1739.

    Google Scholar 

  39. Santana MJ, Feeny D, Jackson K, Weinkauf J, Lien D. Improvement in health-related quality of life after lung transplantation. Can Respir J. 2009;16(5):153–8.

    Article  Google Scholar 

  40. Smeritschnig B, Jaksch P, Kocher A, Seebacher G, Aigner C, Mazhar S, Klepetko W. Quality of life after lung transplantation: a cross-sectional study. J Heart Lung Transplant. 2005;24(4):474–80.

    Article  Google Scholar 

  41. Stącel T, Jaworska I, Zawadzki F, Wajda-Pokrontka M, Tatoj Z, Urlik M, Latos M, Szywacz W, Szczerba A, et al. Assessment of Quality of Life Among Patients After Lung Transplantation: A Single-Center Study. Transplant Proc. 2020;52(7):2165–72.

    Article  Google Scholar 

  42. Kariv G, Shani V, Goldberg E, Leibovici L, Paul M. A model for diagnosis of pulmonary infections in solid-organ transplant recipients. Comput Methods Programs Biomed. 2011;104(2):135–42.

    Article  Google Scholar 

  43. Leppke S, Leighton T, Zaun D, Chen S-C, Skeans M, Israni AK, Snyder JJ, Kasiske BL. Scientific Registry of Transplant recipients: collecting, analyzing, and reporting data on transplantation in the United States. Transplantation Reviews. 2013;27(2):50–6.

    Article  Google Scholar 

  44. Guijo-Rubio D, Gutiérrez PA, Hervás-Martínez C. Machine learning methods in organ transplantation. Curr Opin Organ Transplant. 2020;25(4):399–405.

    Article  Google Scholar 

  45. Balch JA, Delitto D, Tighe PJ, Zarrinpar A, Efron PA, Rashidi P, Upchurch GR Jr, Bihorac A, Loftus TJ. Machine learning applications in solid organ transplantation and related complications. Front Immunol. 2021;12:739728.

    Article  CAS  Google Scholar 

  46. Shahmoradi L, Abtahi H, Amini S, Gholamzadeh M. Systematic review of using medical informatics in lung transplantation studies. Int J Med Informatics. 2020;136:104096.

    Article  Google Scholar 

  47. Getz KD, He C, Li Y, Huang YV, Burstein DS, Rossano J, Aplenc R. Successful merging of data from the United Network for Organ sharing and the Pediatric Health Information System databases. Pediatr Transpl. 2018;22(5):e13168.

    Article  Google Scholar 

  48. Massie AB, Kucirka LM, Segev DL. Big data in organ transplantation: registries and administrative claims. Am J Transplant. 2014;14(8):1723–30.

    Article  CAS  Google Scholar 

  49. Thaler S, Menkovski V. The role of deep learning in improving healthcare. In: Data Science for Healthcare. edn.: Springer; 2019. pp. 75–116.

  50. Subramanian J, Simon R. Overfitting in prediction models – is it a problem only in high dimensions? Contemp Clin Trials. 2013;36(2):636–41.

    Article  Google Scholar 

  51. Alelyani S. Stable bagging feature selection on medical data. J Big Data. 2021;8(1):1–18.

    Article  Google Scholar 

  52. Bagherzadeh-Khiabani F, Ramezankhani A, Azizi F, Hadaegh F, Steyerberg EW, Khalili D. A tutorial on variable selection for clinical prediction models: feature selection methods in data mining could improve the results. J Clin Epidemiol. 2016;71:76–85.

    Article  Google Scholar 

  53. Sutton RT, Pincock D, Baumgart DC, Sadowski DC, Fedorak RN, Kroeker KI. An overview of clinical decision support systems: benefits, risks, and strategies for success. NPJ Digit Med. 2020;3(1):1–10.

    Article  Google Scholar 

  54. Murphy EV. Clinical decision support: effectiveness in improving quality processes and clinical outcomes and factors that may influence success. Yale J Biol Med. 2014;87(2):187.

    Google Scholar 

Download references

Acknowledgements

We would also like to express our gratitude to Farideh Namazi for their support of us during this research.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Author information

Authors and Affiliations

Authors

Contributions

Conception and design of the study: Reza Safdari, Hamidreza Abtahi, Marsa Gholamzadeh; Acquisition of data: Hamidreza Abtahi, Marsa Gholamzadeh, Reza Safdari. Analysis and/or interpretation of data: Marsa Gholamzadeh, Reza Safdari. Drafting the manuscript: Hamidreza Abtahi, Marsa Gholamzadeh. Revising the manuscript critically for important intellectual content: Reza Safdari, Marsa Gholamzadeh. Approval of the version of the manuscript to be published: Reza Safdari, Hamidreza Abtahi, Marsa Gholamzadeh. The author(s) read and approved the final manuscript.

Corresponding author

Correspondence to Reza Safdari.

Ethics declarations

Ethics approval and consent to participate

This study was approved by the ethics committee of the Tehran University of Medical Sciences with an ethical code (IR.TUMS.SPH.REC.1400.340). The authors have no ethical conflicts to disclose.

Consent for publication

Not Applicable.

Competing interests

The authors declare no competing interests.

Additional information

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Gholamzadeh, M., Abtahi, H. & Safdari, R. Machine learning-based techniques to improve lung transplantation outcomes and complications: a systematic review. BMC Med Res Methodol 22, 331 (2022). https://doi.org/10.1186/s12874-022-01823-2

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/s12874-022-01823-2

Keywords

  • Lung transplantation
  • Machine learning
  • Review
  • Lung diseases