What is Predictive Medicine?
• Predictive medicine is a branch of medicine that’s goal is to identify different patients who are at risk for a disease which allow prevention or early treatment of that disease.
What is Predictive Medicine?
• Predictive medicine is a branch of medicine that’s goal is to identify different patients who are at risk for a disease which allow prevention or early treatment of that disease.
• It is a proactive approach that uses different tools, analytics and algorithms to develop solutions to potential health problems
What is Predictive Medicine?
• Predictive medicine is a branch of medicine that’s goal is to identify different patients who are at risk for a disease which allow prevention or early treatment of that disease.
• It is a proactive approach that uses different tools, analytics and algorithms to develop solutions to potential health problems
• The value that predictive medicine provides is it: increases accuracy of diagnoses, the model become more accurate overtime and patients have better outcomes
Why Try To Predict Diabetes?
• In 2012, 29.1 million Americans or 9.3% of the population was reported to have diabetes.
Why Try To Predict Diabetes?
• In 2012, 29.1 million Americans or 9.3% of the population was reported to have diabetes.
• Of those 29.1 million Americans, 21.0 million were diagnosed while 8.1 million were undiagnosed.
Why Try To Predict Diabetes?
• In 2012, 29.1 million Americans or 9.3% of the population was reported to have diabetes.
• Of those 29.1 million Americans, 21.0 million were diagnosed while 8.1 million were undiagnosed.
• The prevalence in seniors above the age of 65 or older remains high, at 25.9%.
Why Try To Predict Diabetes?
• In 2012, 29.1 million Americans or 9.3% of the population was reported to have diabetes.
• Of those 29.1 million Americans, 21.0 million were diagnosed while 8.1 million were undiagnosed.
• The prevalence in seniors above the age of 65 or older remains high, at 25.9%.
• In 2012, 86 million Americans that were over the age of 20 had prediabetes, which is 7 million more than in 2010.
My Process
Remove patients whose age at death is less than their age at diagnosis:
168 patients Original Dataset:
17,443,442 instances
Updated Dataset:
17,432,694 instances
Updated Dataset:
17,379,218 instances
Remove patients who had more than 365
diagnoses in a year: 56 patients
Remove patients whose age of first diagnosis and last diagnosis is
more than a year:158,068 patients
Updated Dataset:
7,002,530 instances
Updated Dataset:
7,002,530 instances
Remove patients whose age at diagnosis is
greater than 110 and do not have an entry for
their Age or icd9:0 patients
Average Age and Standard Deviation
• Average age of patients with diabetes: 60.01• Standard Deviation: 4.33• After these findings:
– Patients with their diagnosis of diabetes before the age of 40 were removed from the dataset.
Updated Dataset:
7,002,530 instances
Final Dataset to use:
6,925,196 instances
Randomization
Final Dataset to use: 6,925,196
instances
Training Set: 6,232,676 instances
Validation Set: 692,520 instances
90% of Data
10% of Data
Calculation of LR
Diabetic/Cases__________
NonDiabetic/ControlsNote: IIF statements were used in correlation to account for values
of 0.Note: with ICD “250”, were
excluded from LR and posterior odds calculation
Top Ten ICD9’s Associated With Diabetes
Name ICD9 Diabetic NonDiabetic Cases Controls LR
Background diabetic retinopathy I362.01 7 1 160579 433915 18.92
Parvovirus B19 I079.83 5 1 160579 433915 13.51
Chronic meningitis I322.2 4 1 160579 433915 10.81
Benign neoplasm of trachea I212.2 4 1 160579 433915 10.81
Malignant histiocytosis I202.3 3 1 160579 433915 8.11
Peripheral angiopathy in diseases classified elsewhere I443.81 3 1 160579 433915 8.11
Traumatic spondylopathy I721.7 3 1 160579 433915 8.11
Open fracture of unspecified part of neck of femur IE820.9 3 1 160579 433915 8.11
Poisoning by psychostimulants I969.7 5 2 160579 433915 6.76
Erythema multiforme I695.1 2 1 160579 433915 5.4
Posterior Odds – Top Ten
ID Odds Probability
179490 3.17 0.76
191036 3.15 0.76
132395 2.7 0.73
140581 2.18 0.69
150695 1.81 0.64
170928 1.8 0.64
160199 1.71 0.63
171511 1.56 0.61
190708 1.16 0.54
135569 1.16 0.54
Sensitivity, Specificity Values
Cutoff (Probability) Specificity 1-Specficity Sensitivity
0 0 1 1
0.05 0.0008 0.999 0.99
0.1 0.01 0.99 0.99
0.15 0.037 0.963 0.98
0.2 0.096 0.904 0.95
0.4 0.98 0.02 0.038
0.8 0.998 0.002 0.0007
0.85 0.998 0.002 0.0006
0.9 0.999 0.001 0.0004
1 1 0 0
Sensitivity vs 1-Specficity
AOC: .58, Accuracy = 58%
Area of Square (0.5)*(0.65) + Triangle#1:
(0.5)*(0.5)*(0.35) + Triangle#2
(0.65)*(0.5)*(0.5) + = .58
Conclusion
• In 2012, 29.1 million Americans or 9.3% of the population was reported to have diabetes.
Conclusion
• In 2012, 29.1 million Americans or 9.3% of the population was reported to have diabetes.
• An accuracy of 58%, slightly above a random prediction accuracy of 50% is validated due to the degree of difficulty in predicting diabetes in patients.
Conclusion
• In 2012, 29.1 million Americans or 9.3% of the population was reported to have diabetes.
• An accuracy of 58%, slightly above a random prediction accuracy of 50% is validated due to the degree of difficulty in predicting diabetes in patients.
• Continuing to promote predictive medicine is key for the communities who have have some level of diabetes by studying medical history in order to manage their health and prevent diabetes.
References
• http://www.diabetes.org/diabetes-basics/statistics/• http://www.nature.com/subjects/predictive-medicine• http://
www.openhealthnews.com/articles/2012/predictive-medicine-health-it-systems
• https://www.elsevier.com/connect/seven-ways-predictive-analytics-can-improve-healthcare