# Smart-EHR
Predicting user's disease by using various Machine Learning models
Introduction
Smart Electronics Health Record was made keeping in mind to ease the medical processes of detecting severe diseases and their treatment. Covering major human diseases like heart, kidney and cancer, we tried to provide an easy, user friendly and near perfect prediction for general people. Through the web based application, a user just by sitting in their homes could know if his/her symptoms will lead to anything or are just temporary.
We have provided an amiable website for users to enter their symptoms and medical values to predict their disease. Just by clicking, one will get to know what is wrong with their heart or lungs or diabetes level. Moreover it’s a age free application i.e. it could be applicable for people of all age groups.
The author of [1] illustrate how healthiness and technology integration and benchmarking have become a challenge for researchers. Being a major factor in strengthening the present medical systems, it also provides a medium to expand the objectives and strategies of organizations.
In [2], it provides a systematic review of the efficacy of patient-oriented care interventions for people with chronic conditions. Around thirty randomized controlled trials were identified from health-related databases. The findings indicated that most interventions were based on the notion of empowering care and included attempts to educate consumers or prompt them about how to manage a health consultation.
The main objective in [3] was to report on the application and accuracy of the popular mining algorithms (artificial neural networks and decision trees) along with logistic regression on a large data set of breast cancer cases, taking advantage of those available technological advancements to develop prediction models for breast cancer survivability.
Research paper [4] intends to provide a survey of current techniques of knowledge discovery in databases using data mining techniques that are in use in today’s medical research particularly in Heart Disease Prediction. It could seen that Decision Tree was in close competition with Bayesian, while algorithms like KNN, Neural Networks, etc were not so efficient. Reduction of actual data size by genetics further improved the efficiency.
In [5], in order to determine how data mining techniques (DMT) and their applications have developed in the past decade, the paper reviews data mining techniques and their applications and development from 2000 to 2011. Keywords were used to identify 216 articles concerning DMT applications, from 159 academic journals. The ability to continually change and acquire new understanding is a driving force for the application of DMT and this will allow many new future applications.
Authors of [6] have provided with the review of recent ML approaches employed in the modeling of cancer progression. The predictive models discussed here are based on various supervised ML techniques as well as on different input features and data samples.
Paper [7] explores the use of machine-learning based alternatives to standard statistical data completion (data imputation) methods, for dealing with missing data. It has used two techniques, mainly, unsupervised clustering strategy which uses a Bayesian approach to cluster the data into classes and modeling missing variables by supervised induction of a decision tree-based classifier.
Technology
Smart Electronics Health Record was made keeping in mind to ease the medical processes of detecting severe diseases and their treatment. Covering major human diseases like heart, kidney and lungs, we tried to provide an easy, user friendly and near perfect prediction for general people. Through the web based application merged with R Programming, a user just by sitting in their homes could know if his/her symptoms will lead to anything or are just temporary.
We have provided an amiable website for users to enter their symptoms and medical values to predict their disease. The website is made with the help of a new technology, namely R Shiny. With a little bit help of HTML, we were able to integrate various diseases under one url.
Our purpose was to find the best algorithm which could work for all the diseases. To do so, we have applied eight different machine learning and deep learning algorithms. These are : Linear, Bayesian, Random Forest, Extreme GB Boosting, Decision Trees, Genetic Algorithms and Neural Networks with Tensor Flow. At the end, we found that Extreme GB Boosting gives the most promising results in all the disease data sets.
Besides Classifications, we have also provided data visualization to show relationship between different attributes for a disease. We have used various packages of R, like ggplots, pie charts, time analysis etc, for a better experience.
Description
Smart EHR covers majorly five diseases: Heart, Diabetes, Parkinson’s Kidney and Breast Cancer. Before going further deep into the technical aspects, it is important to understand what each data sets look like and their attributes.
Heart
The heart dataset has 10 attributes which contribute to the prediction of the disease. Each attribute its own level of importance for the cause of heart disease.
Table 1. Attributes of Heart dataset
S.No Name Description
1. Cp Chest pain type
2. Trestbps Resting blood pressure (in mm Hg on admission to the hospital)
3. Chol Serum cholestoral in mg/dl
4. Fbs Fasting blood sugar
5. Restecg Resting electrocardiographic results
6. Thalach Maximum heart rate achieved
7. Exang Exercise induced angina
8. Op Observered pressure
8. Slope The slope of the peak exercise ST segment
9. Ca Number of major vessels (0-3) colored by flourosopy
10. Thal Defect in Thal level
11. Heart Is there a heart disease or no
Diabetes
Not only insulin levels in the body affects Diabetes, but there are many other factors like body mass index, age etc also impacts on the level of Diabetes.
Table 2. Attributes of Diabetes dataset
S.No Name Description
1. Pregnant Number of times pregnant
2. Plasma_Glucose Plasma glucose concentration in the body
3. Dias_BP Blood pressure
4. Triceps_Skin Triceps skin fold thickness
5. Serum_Insulin Serum insulin
6. BMI Body mass index
7. DPF Pedigree function
8. Age Age of the patient
9. Diabetes Is there Diabetes or not
Cancer
The breast cancer dataset has 11 attributes which contribute to the prediction of the disease. Each attribute contributes into the likeliness of the disease.
Table 3. Attributes of Cancer dataset
S.No Name Description
1. diagnosis Diagnosis (M = malignant, B = benign)
2. radius Distance from centre to points on the perimeter
3. texture Standard deviation of grey-scale values
4. perimeter Perimeter
5. area Area
6. smoothness Local variation in radius lengths
7. compactness Perimeter^2 / area - 1.0
8. concavity Severity of concave portions of the contour
9. concave points Number of concave portions of the contour
10. symmetry Symmetry
11. fractal dimension “Coastline approximation" - 1
Parkinson’s
The Parkinson’s dataset has 25 attributes which contribute to the prediction of the disease. Each attribute has its unique importance and fundamental frequency.
Table 4. Attributes of Parkinson’s dataset
S.No Name Description
1. MDVP:Fo(Hz) Average vocal fundamental frequency
2. MDVP:Fhi(Hz) Maximum vocal fundamental frequency
3. MDVP:Flo(Hz) Minimum vocal fundamental frequency
4. MDVP:Jitter(%), MDVP:Jitter(Abs),
MDVP:RAP, MDVP:PPQ, Jitter:DDP Several measures of variation in fundamental frequency
5. MDVP:Shimmer,
MDVP:Shimmer(dB),
Shimmer:APQ3,
Shimmer:APQ5,
MDVP:APQ,
Shimmer:DDA Several measures of variation in amplitude
6. NHR, HNR Two measures of ratio of noise to tonal components in the voice
7. Status Health status of the subject (one) - Parkinson's, (zero) - healthy
8. RPDE, D2 Two nonlinear dynamical complexity measures
9. DFA Signal fractal scaling exponent
10. spread1, spread2, PPE Three nonlinear measures of fundamental frequency variation
Kidney
The kidney dataset has 25 attributes which contribute to the prediction of the disease, which includes parameters like Red Blood Cells, Pus cells clumps, Appetite and so on.
Table 5. Attributes of Kidney Dataset
S.No Name Description
1. age Age
2. bp Blood pressure
3. sg Specific gravity
4. al Albumin
5. su Sugar
6. rbc Red blood cells
7. pc Pus cell
8. pcc Pus cell clumps
9. ba Bacteria
10. bgr Blood glucose random
11. bu Blood urea
12. sc Serum creatinine
13. sod Sodium
14. pot Potassium
15. Hemo Hemoglobin
16. pcv Packed cell volume
17. wc White blood cell count
18. rc Red blood cell count
19. Htn Hypertension
20. Dm Diabetes mellitus
21. cad Coronary artery disease
22. appet Appetite
23. Pe Pedal edema
24. Ane Anaemia
25. Class Class variable 0 – not CKD 1- CKD
Classification
Classification is the technique which classifies the whole data set according to the class variable. To maintain the credibility of the prediction, we incorporated seven machine learning algorithms for classifying the data set.
Logistic Regression
Logistic regression is a classification regression model where the dependent variable is categorical and is needed to be classified in binary classes. The case of a binary dependent variable that is, where the output can take only two values, "0" and "1", which represent outcomes such as pass/fail, win/lose, alive/dead or healthy/sick. Logistic regression is similar to linear regression but unlike it, logistic regression is a classification algorithm. In the project we used the caret module to implement the logistic regression on the given data to predict and classify the tests.
Logistic regression simply uses the technique of reducing the error function and hence deducing the equation for the line of predicted values as in linear regression but it uses a damping function to contain the values between 1 and 0. In general it proved to be a very accurate algorithm as it gave accuracy of 75% in all the datasets that we used. Its average accuracy was 78.3%.
Bayesian Logistic Regression
Bayesian analyses of binary or categorical outcomes typically rely on probability or mixed effects logistic regression models which do not have a marginal logistic structure for the individual outcomes. The BLR model for individual outcomes has a marginal logistic structure, simplifying interpretation and follow a Bayesian approach to estimation and inference, developing an efficient data augmentation algorithm for computation.
Bayesian logistic regression provide with accurate results of around 80%. It is a very efficient algorithm especially in the parkinson’s dataset in which it gave an accuracy of 87%.
Decision Trees
A decision tree is a flowchart-like structure in which each internal node represents a "test" on an attribute , each branch represents the outcome of the test, and each leaf node represents a class label. The paths from root to leaf represent classification rules.
Decision tree is a classification technique that is easy to interpret and understand and hence easy to use. It can be used in both normal and hard data and is pretty accurate if the data is not overfitting to the classifiers.
Random Forest
Random forest is a tree-based machine learning algorithm which involves building several trees and combining their output to predict outcomes.
Firstly, the dataset is split into random samples by using bootstrap aggregating algorithm. A new dataset is created by sampling n random cases from original data leaving about one third of data.
Then the new data obtained is trained using the given model in which very small columns of data are selected at random. At the end, each selected data is allowed to grow fully and final prediction is obtained by averaging.
Extreme Gradient Boosting
Gradient boosting is a machine learning algorithm which uses weak prediction models to create a final prediction model.
At first data is modelled onto simple models like decision trees and check for error residuals.
For further models we fit the data onto error residuals as target with same input variables.
Now it add the new error residuals with the original ones and fit the data onto updated error residuals.
Repetition of the steps are done until the sum of residuals become constant. In the end all predictors are added to predict outcome.
Genetic Algorithm
Genetic algorithms are search based algorithms based on natural selection, recombination and mutation to evolve solutions to a problem.
First, we have a pool of different solutions in hand. These solutions undergo recombination and mutation, and in process produces new solutions. Each solution is given a fitness score accordingly.
Crossover is performed by randomly selecting solutions to produce offspring on the basis of fitness score. The fitter the solutions are, the fitter is their offspring. The process is repeated till a new batch of fitter solutions are created.
Neural Networks
Artificial Neural Networks work similar to that of a Human Brain. It has the ability to learn on its own from the dataset, so as to make accurate and precise prediction and classification.
Neural Networks have hidden layer in order to increase the productivity and its learning rate. In the first level, input is taken from the parameters itself to compute prediction. This level forms the first hidden layer for Neural Networks.
It then learns from the output about the error rate and recursively compute other predictions to give us a final answer, which has minimum error rate or high accuracy percentage.
The more the number of hidden layer, the more efficiently the model could learn on its own. Thus Neural Network is one of the finest and commonly used Machine Learning algorithms, used for unfeigned classification and prediction.
Comparison Models
After applying seven algorithms i.e. Linear Regression, Baeysian Regression, Decision Tree, Random Forest, Extreme Gradient Boosting, Genetic Algorithm and Neural Networks. on all the 5 datasets and comparing their accuracies we came to a conclusion that Extreme Gradient Boosting is giving the best accuracy on all datasets. Hence we have proceeded using by extreme gradient boosting algorithm for predicting the outcomes. Other algorithms were giving varying results in different datasets, thus convincing us not to rely on them.
Extreme Gradient Boosting, though not giving hundred percent accuracy, but was reliable. It gave considerable results with around ninety percent accuracy for each disease dataset. Hence we chose Extreme Gradient Booster as the optimal algorithm for Smart Electronics Health Record.
WEB BASED APPLICATION
We incorporated a responsive web based frontend for easier and better experience for the users. They could easily enter the values of various parameters and see the output, thus taking steps accordingly.
The web based application was made with R Shiny. The UI is highly responsive i.e. changing value of the single parameter will enable the computation again.
Users have the choice to test any of the disease provided that are : Heart, Diabetes, Kidney, Parkinson and Cancer. They could easily shuffle through tabs to enter their parameters value, thus giving them an user friendly web experience.
REFERENCES
With the help and guidance of many people and online websites, we were able to achieve what we wished. Some of the references are:
[1] Health Information Systems. Understanding Health Care IT Alignment, (PMID:20938567), Bréant C, Yearbook of medical informatics [2010;:30-3]
[2] McMillan, S. S., Kendall, E., Sav, A., King, M. A., Whitty, J. A., Kelly, F., & Wheeler, A. J. (2013). Patient-centered approaches to health care: A systematic review of randomized controlled trials.
[3] Predicting breast cancer survivability: a comparison of three data mining methods, Author Dursun Delen, Glenn Walker, Amit Kadam, Department of Management Science and Information Systems, Oklahoma State University, 700 North Greenwood Venue, Tulsa, OK 74106, USA
[4] Predictive Data Mining for Medical Diagnosis: An Overview of Heart Disease Prediction, Jyoti Soni, Ujma Ansari, Dipesh Sharma, Sunita Soni, International Journal of Computer Applications (0975 – 8887) Volume 17– No.8, March 2011
[5] Data mining techniques and applications – A decade review from 2000 to 2011, Sh HsienLiao, Pei-HuiChu, Pei-YuanHsiao, Department of Management Sciences, Tamkang University, No. 151, Yingzhuan Rd., Tamsui Dist., New Taipei City 25137, Taiwan, ROC
[6] Machine learning applications in cancer prognosis and prediction, Konstantina Kourou, Themis P.Exarchos, , Konstantinos P.Exarchos, , Dimitrios I.Fotiadis, Unit of Medical Technology and Intelligent Information Systems, Dept. of Materials Science and Engineering, University of Ioannina, Ioannina, Greece, IMBB — FORTH, Dept. of Biomedical Research, Ioannina, Greece, Molecular Oncology Unit, Department of Biological Chemistry, Medical School, University of Athens, Athens, Greece
[7] Imputation of missing data using machine learning techniques Kamakshi Lakshminarayan, Steven A. Harp, Robert Goldman and Tariq Samad 3660 Technology Drive Honeywell Technology Center Minneapolis, MN55418, USA
[8] Nikunj Agarwal, M.P.Sebastian, “Use of Cloud Computing and Smart Devices in Healthcare”, WASET, Intl. Jr. Computer, Electrical, Automation, Control and Information Engineering, Vol.10(1): 156-159, 2016.
[9] "Obesity In Australia Modi", Modi.monash.edu.au. N.p 2015, June 2015.
[10] Ya-Li Zheng, Xiao-Rong Ding, Carmen Chung Yan Poon, Unobtrusive Sensing and Wearable Devices for Health Informatics.
[11] Asha Rajkumar, G.Sophia Reena, Diagnosis Of Heart Disease Using Datamining Algorithm, Global Journal of Computer Science and Technology 38 Vol. 10 Issue 10 Ver. 1.0 September 2010.
[12] Sunita Soni, O.P.Vyas, Using Associative Classifiers for Predictive Analysis in Health Care Data Mining, International Journal of Computer Application (IJCA, 0975 – 8887) Volume 4– No.5, July 2010, pages 33-34.
[13] ] W.Li, J. Han, J.Pei , CMAR- Classification based on Multiple Association Rules, ICDM‟01, , San Jose, CA, Nov. 2001. pp. 369-376
[14] Carloz Ordonez, Association Rule Discovery with Train and Test approach for heart disease prediction, IEEE Transactions on Information Technology in Biomedicine, Volume 10, No. 2, April 2006.pp 334-343.
[15] M.Y.C. Polley, B. Freidlin, E.L. Korn, B.A. Conley, J.S. Abrams, L.M. McShaneStatistical and practical considerations for clinical evaluation of predictive biomarkers, J Natl Cancer Inst, 105 (2013), pp. 1677-1683
[16] O. Fortunato, M. Boeri, C. Verri, D. Conte, M. Mensah, P. Suatoni, et al.Assessment of circulating microRNAs in plasma of lung cancer patients, Molecules, 19 (2014), pp. 3038-3054
[17] Rothman, Brian; Joan. C. Leonard; Michael. M. Vigoda (2012). "Future of electronic health records: implications for decision support". Mount Sinai Journal of Medicine 79 (6): 757-768.
[18] J. Sun, C. D. McNaughton, P. Zhang, A. Perer, A. Gkoulalas-Divanis,J. C. Denny, J. Kirby, T. Lasko, A. Saip, and B. A. Malin, “Predicting changes in hypertension control using electronic health records from a chronic disease management program,” J. Amer. Med. Informat. Assoc., vol. 21, pp. 337–344, 2014.
[19] D. W. Bates, S. Saria, L. Ohno-Machado, A. Shah, and G. Escobar, “Big data in health care: Using analytics to identify and manage high risk and high-cost patients,” Health Affairs, vol. 33, pp. 1123–1131, 2014
[20] D. Yach, C. Hawkes, C. L. Gould, K. J. Hofman, "The global burden of chronic diseases: Overcoming impediments to prevention and control", JAMA, vol. 291, no. 21, pp. 2616-2622, 2004.
[21] B. Kayyali, D. Knott, S. Van Kuiken, "The big-data revolution in US health care: Accelerating value and innovation", McKinsey Company, pp. 1-13, 2013.
[22] Heart Disease Facts, 2015, [online] Available: http://www.cdc.gov/heartdisease/facts.htm
[23] Diabetes Globally, 2017, [online] Available: https://www.diabetesaustralia.com.au/diabetes-globally.
[24] D. A. Ludwick, J. Doucette, "Adopting electronic medical records in primary care: Lessons learned from health information systems implementation experience in seven countries", Int. J. Med. Inf., vol. 78, no. 1, pp. 22-31, 2009.
[25] J.Archenaa and E.A. Mary Anita, “A Survey Of Big Data Analytics in Healthcare and Government”, 2nd International Symposium on Big Data and Cloud Computing (ISBCC !"#$% &'()*+,-% ,. Computer Science 50, 408 – 413, 2015.
[26] MimohOjha, KirtiMathur, “Proposed Application of Big Data Analytics in Healthcare at Maharaja Yeshwantrao Hospital”, 2016 3rd MEC International Conference on Big Data and Smart City, IEEE, 2016.