Application of the Intelligent Techniques in Transplantation
Databases: A Review of Articles Published in 2009 and 2010
F.S. Sousa, A.D. Hummel, R.F. Maciel, F.M. Cohrs, A.E.J. Falcão, F. Teixeira, R. Baptista, F. Mancini,
T.M. da Costa, D. Alves, and I.T. Pisa
ABSTRACT
The replacement of defective organs with healthy ones is an old problem, but only a few
years ago was this issue put into practice. Improvements in the whole transplantation
process have been increasingly important in clinical practice. In this context are clinical
decision support systems (CDSSs), which have reflected a significant amount of work to
use mathematical and intelligent techniques. The aim of this article was to present
consideration of intelligent techniques used in recent years (2009 and 2010) to analyze
organ transplant databases. To this end, we performed a search of the PubMed and
Institute for Scientific Information (ISI) Web of Knowledge databases to find articles
published in 2009 and 2010 about intelligent techniques applied to transplantation
databases. Among 69 retrieved articles, we chose according to inclusion and exclusion
criteria. The main techniques were: Artificial Neural Networks (ANN), Logistic
Regres-sion (LR), DeciRegres-sion Trees (DT), Markov Models (MM), and Bayesian Networks (BN).
Most articles used ANN. Some publications described comparisons between techniques or
the use of various techniques together. The use of intelligent techniques to extract
knowledge from databases of healthcare is increasingly common. Although authors
preferred to use ANN, statistical techniques were equally effective for this enterprise.
T
HE replacement of defective organs with healthy ones is an old problem, but only within a few years has this issue been put into practice.1Advances in organ transplan-tation have increased quality of life and survival of patients. In addition, the factor of insufficient supply, high cost, and waiting list mortality make the field of organ transplanta-tion a complex scenario. Therefore, improvements through-out the process have been increasingly important to clinical practice. Improvements not yet mapped in transplantation databases using new data analysis techniques may contrib-ute directly or indirectly to the generation of knowledge.In this context are Clinical Decision Support Systems (CDSSs) defined by Berner as “a computer-based algorithm that assists a clinician with one or more component steps of the diagnostic process.”2The algorithm cooperates with the discovery of knowledge, sometimes surpassing even the difficulties inherent in the nature of medical knowledge. For example, CDSSs can be used as tools identify the severity of diseases and better treatments by helping physicians to make the best decision for each case.2,3
The literature includes applications of intelligent and mathematical techniques to the development of CDSSs
with statistical analysis, mathematical models, and artificial intelligence techniques,4 –7some of them which have been applied in transplantation.8
We sought to present evidence from the specialized scientific literature in which computational techniques were used in 2009 and 2010 to analyze organ transplantation databases.
METHODS
We performed a search of Pubmed (http://www.ncbi.nlm.nih.gov/ pubmed/) and Institute for Scientific Information (ISI) Web of
From the Programa de Pós-graduação em Informática em Saúde (F.S.S., A.D.H., A.E.J.F., F.T., R.B., F.M., T.M.da C.), Universidade Federal de São Paulo, São Paulo, Brazil; Programa de Pós-graduação em Saúde Coletiva (R.F.M., F.M.C.), Depar-tamento de Medicine Social (D.A.), Faculdade de Medicina de Ribeirão Preto, Universidade de São Paulo, Ribeirão Preto-SP, Brazil; Departamento de Informática em Saúde, (I.T.P.) Univer-sidade Federal de São Paulo, Brazil.
Address reprint requests to Ivan Torres Pisa, Rua Botucatu, 862, Vila Clementino, São Paulo, SP, Brasil, CEP: 04023-900. E-mail:buscasaude@unifesp.br
0041-1345/11/$–see front matter © 2011 by Elsevier Inc. All rights reserved. doi:10.1016/j.transproceed.2011.02.028 360 Park Avenue South, New York, NY 10010-1710
Knowledge (http://apps.isiknowledge.com) databases. Our inclu-sion criteria considered articles containing the search descriptors present in the title, abstract, keywords, and subject in 2009 and 2010. As exclusion criteria, we considered: the absence of a full text, a topic unrelated to transplantation databases or application of mathematical methods, and documents that were not original, such as literature reviews, letters, and comments. The searches used are as follows.
Pubmed: (“Transplantation”[MESH] AND (“artificial intelli-gence”[MESH] OR “Decision support systems clinical”[MESH]) OR “Neural Networks (Computer)”[MESH]) OR (“Decision Trees/methods”[MESH] OR “Decision Trees/statistics and numer-ical data”[MESH]) OR “Fuzzy Logic”[MESH]) OR “Decision Support Techniques”[MESH])) AND “2009” [Publication Date]: “2010”[Publication Date].
ISI Web of Knowledge: Topic⫽(“Transplantation” AND
(“arti-ficial intelligence” OR “Decision support systems” OR “Arti(“arti-ficial Neural Networks” OR “Decision Trees” OR “Fuzzy Logic” OR “Decision Support Techniques”)) AND Year Published⫽(2009
OR 2010).
RESULTS
We found 69 articles including 2 common to both data-bases. We excluded 59 of these due to the above exclusion criteria, leaving 8 files for reading. There were 3 more articles in 2010 than in previous work.8 The following techniques were identified: Artificial Neural Networks (ANN), Logistic Regression (LR), Decision Trees (DT), Markov Models (MM), and Bayesian Networks (BN). Table 1summarizes the applications and techniques used in these articles.
Five of the 8 articles described the use of ANN. Nilsson et al9,10 developed a predictive model for survival and donor-recipient matching using ANN. Using a database of 70,759 records from heart transplantation,9 they selected the most important variables, achieving a correlation of 0.59 between ANN and Kaplan-Meier results. In a database of 59,698 records of heart transplantations10 as well, the model selected 14 donor and 21 recipient important vari-ables, observing that the best donor-recipient matching achieved a survival 5-year-transplant rate of 77%.
Hummel et al11applied ANN to a database of 145 kidney transplant recipients seeking to predict nephrotoxicity and acute cellular rejection (ACR) after the first year posttrans-plantation. For nephrotoxicity, the authors achieved
accu-racy (ACC) of 75.68%, whereas for ACR, it reached 80.89%.
Oztekin et al12compared ANN, LR, and DT applied to a database of 16,604 heart-lung transplantation records. They showed that ANN and LR achieved similar results to predict survival at 9 years with accuracies of 82.4% and 81.9% respectively. These methods showed better perfor-mances than DT (ACC⫽74.9%). Similarly, Caocci et al13 showed that ANN outperformed LR to predict the occur-rence of acute graft-versus host disease (aGVHD) among a database of 78 stem cell transplant recipients with an accuracy 83.3%.
Using a more probabilistic approach, Pidala et al14 employed MM to predict the optimal type of stem cells for transplantation (peripheral blood or bone marrow), based upon estimated survival after transplantation. The survival curves produced by the model were similar to those ob-served by meta-analysis. The authors also found that the life expectancy of those who receive peripheral blood stem cells was in most cases greater than that of those who received bone marrow stem cells. Also using MM, Saab et al15 compared the cost-effectiveness of 2 models of hepatitis B prevention in liver transplant recipients.
Finally, Stajduhar et al16 compared several BN algo-rithms among a database of 137 patients seeking to predict relapse after bone marrow transplantation. They observed which variables were important for this type of classifica-tion, achieving an accuracy rate of 77.43%.
DISCUSSION
The use of ANN seems to be preferred for knowledge extraction from transplantation databases. Consistent with the review of Hummel et al,8we noted the application of ANN in most articles. Some of these compared ANN with other learning techniques; ANN outperformed DT12and LR.13Moreover, ANNs were used to predict survival,9,10,12 risk of a bad match between donor and recipient,9,10 occurrence of diseases after transplantation,13and rejection or nephrotoxicity.11It can also be used in conjunction with other techniques such as simulations based on Monte Carlo models.10 Despite the good results presented in these publications, one must be careful to not generalize them. Many articles showed results of applications to specific databases, which were used only during training, while some
Table 1. Summary of Application and Techniques Used in Found Articles
Article Application Technique
Nilsson et al (2009) To predict survival after heart transplantation ANN
Oztekin et al (2009) To improve results after heart-lung transplantation ANN LR, and DT Pidala et al (2009) To decide the type of stem cells to transplant MM
Saab et al (2009) To compare cost-effectiveness of 2 models to prevent hepatitis B after liver transplantation
MM
Stajduhar et al (2009) To predict relapse after bone marrow transplantation BN Caocci et al (2010) To predict aGVHD after stem cells transplantation ANN and LR Hummel et al (2010) To determine nephrotoxicity and ACR after kidney transplantation ANN Nilsson et al (2010) To predict survival after heart transplantation ANN
used databases with few data. When used in clinical practice or in various databases, the results can be different. Only Stajduhar et al16extrapolated their methodology to differ-ent databases, achieving significant results in all of them.
It is worthwhile to emphasize the wide applicability of mathematical and computational techniques. We can find applications to different types of transplantation such as bone marrow, heart, lung, and kidney. Hummel et al8also found applications in previous studies of liver and pediatric transplantation.
In conclusion, the use of intelligent techniques to extract knowledge from transplantation databases is increasingly common. The preference for and efficacy of ANN is evi-dent. Moreover, statistical techniques such as MM have proven effective equally to achieve the proposed objectives. It is clear that mathematical and computational techniques can work well with transplantation databases. We hope that this work will contribute to increase researchers interest in applying intelligent techniques to knowledge discovery in organ transplant databases and serve as a guide to choose the best techniques to apply in their problems.
REFERENCES
1. Kuss R, Bourge P: An illustrated history of organ transplan-tation: the great adventure of the century. Rueil-Malmaison, France: Laboratoires Sandoz; 1992, p. 10
2. Berner ES: Clinical decision support systems: theory and practice. New York: Springer; 1999,
3. Bouchon-Meunier B: Uncertainty management in medical applications. In Nonlinear Biomedical Signal Processing: Fuzzy Logic, Neural Networks, and New Algorithms. New York: IEEE Press; 2000, p 1
4. Banerjee R, Das A, Ghoshal UC, et al: Predicting mortality in patients with cirrhosis of liver with application of neural network technology. J Gastroenterol Hepatol 18:1054, 2003
5. Marinho A: A study on organ transplantation waiting lines in Brazil’s Unified National Health System. Cadernos de Saúde Pública 22:2229, 2006
6. Murugan R, Venkataraman R, Wahed AS, et al: HIDonOR Study Investigators. Increased plasma interleukin-6 in donors is associated with lower recipient hospital-free survival after cadav-eric organ transplantation. Crit Care Med 36:1810, 2008
7. Patel S, Cassuto J, Orloff M, et al: Minimizing morbidity of organ donation: analysis of factors for perioperative complications after living-donor nephrectomy in the United States. Transplanta-tion 85:561, 2008
8. Hummel AD, Maciel RF, Falcão AE, et al: Aplicação de Técnicas Computacionais em Bases de Da-dos de Transplante: Revisão de Artigos Publicados no Bie˜nio 2007–2008. JBT Jornal Brasileiro de Transplantes 12:1045, 2009
9. Nilsson J, Ohlsson M, Hoglund P, et al: Virtual cross-matching in heart transplantation: large scale simulation of survival after heart transplantation using artificial neural networks. J Heart Lung Transplant 28(suppl 1):S231, 2009
10. Nilsson J, Ohlsson M, Hoglund P, et al: Artificial neural networks - a method for optimal donor-recipient matching. Large scale simulation of survival after heart transplantation. J Heart Lung Transplant 29(suppl 1):S29, 2010
11. Hummel AD, Maciel RF, Rodrigues RGS, et al: Application of artificial neural networks in renal transplantation: classification of nephrotoxicity and acute cellular rejection episodes. Transplant Proc 42:471, 2010
12. Oztekin A, Delen D, Kong ZJ: Predicting the graft survival for heart-lung transplantation patients: an integrated data mining methodology. Int J Med Inform 78:e84, 2009
13. Caocci G, Baccoli R, Vacca A, et al: Comparison between an artificial neural network and logistic regression in predicting acute graft-vs-host disease after unrelated donor hematopoietic stem cell transplantation in thalassemia patients. Exp Hem 38:426, 2010
14. Pidala J, Anasetti C, Kharfan-Dabaja MA, et al: Decision analysis of peripheral blood versus bone marrow hematopoietic stem cells for allogeneic hematopoietic cell transplantation. Bio Blood Marrow Transplant 15:1415, 2009
15. Saab S, Ham MY, Stone MA, et al: Decision analysis model for hepatitis B prophylaxis one year after liver transplantation. Liver Transplant 15:413, 2009
16. Stajduhar I, Dalbelo-Basic B, Bogunovic N: Impact of censoring on learning Bayesian networks in survival modelling. Artif Intell Med 47:199, 2009
