Location
Nessmith-Lane Atrium
Session Format
Poster Presentation
Research Area Topic:
Public Health & Well Being - Epidemiologic Research
Co-Presenters and Faculty Mentors or Advisors
Lili Yu (Georgia Southern University)
Abstract
Introduction: Acute myeloid leukemia (AML) and acute lymphoblastic leukemia (ALL) are two types of acute leukemia. When complete remission of leukemia has not been achieved or the disease refracts to its stage in initial chemotherapy, relapse occurs. Leukemia has poor prognosis for patients with relapse. This sample included AML and ALL patients. The purpose of this study was to use and extension of the Expectation-Maximization (EM) algorithm in order to find significant factors that affect the occurrence of relapse in leukemia patients with missing data.
Methods: The EM logistic model consists of three steps. First, the initial logistic regression intercept and coefficients are estimated using the complete data, the data with relapse outcome present. Next, the predicted probability of relapse is calculated based on the number of patients relapsed from the complete data. This probability is then used to determine relapse outcome (Y = 0 or 1) in the patients with missing data in the expectation step. During the expectation step, predicted probabilities are produced based on the logistic regression model. We proposed using the mean of the predicted probabilities as the cut-off point for determining whether the missing binary values are set equal to 1 or 0 during imputation. In the maximization step, the logistic regression intercept and coefficients are updated until the estimates converge.
Results: The results indicate that there are a number of significant variables associated with leukemia relapse including the sex of the donor, patient cytomegalovirus status, and FAB (French-American-British classification of AML) grade. Conclusion/Discussion: This study used the EM algorithm and the mean predicted probabilities as the cut-off point during imputation to predict which factors can affect relapse outcome in leukemia patients who had received a bone marrow transplant. Public health significance: The EM algorithm can improve leukemia treatment outcomes. By using the EM algorithm in conjunction with selecting optimal cut points for imputation of a dependent categorical variable using the AUC, public health professionals can find which factors are associated with relapse in order to prevent relapse by controlling for these factors. They can also use the EM algorithm to predict relapse and recommend treatment plans for different patients.
Keywords
Georgia Southern University, Research Symposium, Leukemia patients, Missing data, EM algorithm
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Presentation Type and Release Option
Presentation (Open Access)
Start Date
4-16-2016 10:45 AM
End Date
4-16-2016 12:00 PM
Recommended Citation
Schaible, Braydon, "Analysis of Relapse in Leukemia Patients With Missing Data Using an Extension of the EM Algorithm" (2016). GS4 Georgia Southern Student Scholars Symposium. 6.
https://digitalcommons.georgiasouthern.edu/research_symposium/2016/2016/6
Included in
Analysis of Relapse in Leukemia Patients With Missing Data Using an Extension of the EM Algorithm
Nessmith-Lane Atrium
Introduction: Acute myeloid leukemia (AML) and acute lymphoblastic leukemia (ALL) are two types of acute leukemia. When complete remission of leukemia has not been achieved or the disease refracts to its stage in initial chemotherapy, relapse occurs. Leukemia has poor prognosis for patients with relapse. This sample included AML and ALL patients. The purpose of this study was to use and extension of the Expectation-Maximization (EM) algorithm in order to find significant factors that affect the occurrence of relapse in leukemia patients with missing data.
Methods: The EM logistic model consists of three steps. First, the initial logistic regression intercept and coefficients are estimated using the complete data, the data with relapse outcome present. Next, the predicted probability of relapse is calculated based on the number of patients relapsed from the complete data. This probability is then used to determine relapse outcome (Y = 0 or 1) in the patients with missing data in the expectation step. During the expectation step, predicted probabilities are produced based on the logistic regression model. We proposed using the mean of the predicted probabilities as the cut-off point for determining whether the missing binary values are set equal to 1 or 0 during imputation. In the maximization step, the logistic regression intercept and coefficients are updated until the estimates converge.
Results: The results indicate that there are a number of significant variables associated with leukemia relapse including the sex of the donor, patient cytomegalovirus status, and FAB (French-American-British classification of AML) grade. Conclusion/Discussion: This study used the EM algorithm and the mean predicted probabilities as the cut-off point during imputation to predict which factors can affect relapse outcome in leukemia patients who had received a bone marrow transplant. Public health significance: The EM algorithm can improve leukemia treatment outcomes. By using the EM algorithm in conjunction with selecting optimal cut points for imputation of a dependent categorical variable using the AUC, public health professionals can find which factors are associated with relapse in order to prevent relapse by controlling for these factors. They can also use the EM algorithm to predict relapse and recommend treatment plans for different patients.
