Algorithms to predict clinical response, adherence, and shunting with thiopurines

• US 8,126,690 B2
• Filed: 05/18/2007
• Issued: 02/28/2012
• Est. Priority Date: 05/18/2007
• Status: Active Grant

• Current Assignee: University of Michigan
• Original Assignee: Board of Regents of the University of Michigan (University of Michigan)
• Inventors: Higgins, Peter, Zhu, Ji, Waljee, Akbar, Wang, Sijian, Joyce, Joel
• Primary Examiner(s): BORIN, MICHAEL L
• Application Number:
  US11/804,366
• Publication Number:
  US 20080288227A1
• Time in Patent Office: 1,747 Days
• Field of Search: None
• US Class Current: 703/11
• CPC Class Codes: G16H 50/20 for computer-aided diagnosis, e.g. based on medical expert systems

Abstract

A method of using a variable set from complete blood counts and blood chemistry panels to generate a machine learned algorithm for determining the effectiveness of thiopurine treatment on inflammatory bowel disease (IBD) patients using CART, boosted trees, random forest classification, RuleFit and/or logistic regression analysis.

Claims

1. A method of determining the effectiveness of thiopurine treatment in a patient receiving the thiopurine treatment for inflammatory bowel disease (IBD) comprising:

receiving on a computer device a set of data inputs related to characteristics of a patient receiving a thiopurine treatment for inflammatory bowel disease; and

calculating on the computer device a clinical response probability based on the set of data inputs using an algorithm that applies a weight to each data input of the set of data inputs, wherein each weight relates to a quantified importance of a particular data input associated with a fixed set of patient data from patients receiving a thiopurine treatment for inflammatory bowel disease in which the clinical response of the patients is known,

wherein the set of data comprises at least three data inputs selected from the group consisting of hemoglobin (HGB), hematocrit (HCT), platelet count (PLT), red blood cell count (RBC), mean corpuscular hemoglobin (MCH), mean corpuscular hemoglobin concentration (MCHC), red cell distribution width (RDW), mean platelet volume (MPV), sodium level (SOD), potassium level (POT), chloride (CHLOR), carbon dioxide (CO2), urea nitrogen (UN), creatinine level (CREAT), glucose level (GLUC), calcium level (CAL), protein level (PROT), albumin level (ALB), aspartate transaminase (AST), alanine transaminase (ALT), alkaline phosphatase (ALK), bilirubin total (TBIL), neutrophil count (NEUT), monocyte count (MONO), basophil count (BASO), and age of patient (Age).

2. The method of claim 1, wherein the set of data inputs include at least four of red blood cell volume distribution width (RDW), eosinophil count (EOS), albumin level (ALB), alkaline phosphatase level (ALK), sodium level (SOD), mean corpuscular hemoglobin content (MCH), neutrophil count (NEUT), platelet count (PLT), creatinine level (CREAT), hematocrit (HCT), or age of patient (AGE).

3. The method of claim 1, wherein the probability is calculated using an algorithm generated by a modeling analysis selected from the group consisting of a logistic regression analysis, a classification and regression tree (CART) analysis, a random forest analysis, RuleFit, and a boosted tree regression.

4. The method of claim 1, wherein determining the clinical response probability is calculated based on a statistical regression model, and the set of data inputs for calculating the clinical response probability includes white blood cell count (WBC), red blood cell volume distribution width (RDW), eosinophil count (EOS), albumin level (ALB), alkaline phosphatase level (ALK), sodium level (SOD), and age of patient.

5. The method of claim 1, wherein determining the adherence probability is calculated based on a model derived from a logistic regression using white blood cell count (WBC), mean corpuscular hemoglobin content (MCH), neutrophil count (NEUT), red blood cell volume distribution width (RDW), and age of patient.

6. The method of claim 1, wherein determining the shunting probability is calculated based on a model derived from a logistic regression using platelet count (PLT), red blood cell volume distribution width (RDW), eosinophil count (EOS), albumin level (ALB), creatinine level (CREAT), lymphocyte count (LYMPH), mean corpuscular hemoglobin content (MCH), and hematocrit (HCT).

8. The method of claim 1, wherein the adherence probability is calculated as

probability=1/(1+exp(−

[constant+coefficient1(white blood cell count)+coefficient2(age of patient)+coefficient3(mean corpuscular hemoglobin content)+coefficient4(neutrophil count)+coefficient5(red blood cell volume distribution width)])),

wherein coefficient1 is about 1.756,

coefficient2 is about 1.049,

coefficient3 is about 0.666,

coefficient4 is about 0.578, and

coefficient5 is about 0.656.

9. The method of claim 1, wherein the shunting probability is calculated as

probability=1/(1+exp(−

[constant+coefficient1(eosinophil count)+coefficient2(creatinine level)+coefficient3(lymphocyte count)+coefficient4(mean corpuscular hemoglobin content)])),

wherein coefficient1 is about 0.104,

coefficient2 is about 0.059,

coefficient3 is about 1.518, and

coefficient4 is about 1.149.

7. A method of determining the effectiveness of thiopurine treatment in a patient receiving the thiopurine treatment for inflammatory bowel disease (IBD) comprising:

receiving on a computer device a set of data inputs related to characteristics of a patient receiving a thiopurine treatment for inflammatory bowel disease; and

calculating on the computer device a clinical response probability based on the set of data inputs using an algorithm that applies a weight to each data input of the set of data inputs, wherein each weight relates to a quantified importance of a particular data input associated with a fixed set of patient data from patients receiving a thiopurine treatment for inflammatory bowel disease in which the clinical response of the patients is known,

wherein the set of data inputs comprises white blood cell count, red blood cell volume distribution width, eosinophile count, albumin level, and age patent, and

wherein the clinical response probability is calculated as

probability=1/(1+exp(−

[constant+coefficient1(white blood cell count)+coefficient2(red blood cell volume distribution width)+coefficient3(eosinophil count)+coefficient4(albumin level)+coefficient 5(age of patient)])),

wherein coefficient1 is about 0.754,

coefficient2 is about 0.804,

coefficient3 is about 8.051,

coefficient4 is about 2.903, and

coefficient5 is about 0.972.