althea

ALgoriTHms Exposed through a RESTful API

althos: "healing"

Purpose

The purpose of this application is to make exposing algorithms via a RESTful API easier. After a clinical risk algorithm has been vetted, there are potentially many consumers of this algorithm. For instance, other researchers may be interested in coming up with new state-of-the-art algorithms, clinical operations may wish to use the algorithm to identify cohorts for studies, or in the case of computable phenotypes, real-time event detection. By exposing the algorithm via a RESTful API, we enable all consumers of the algorithm(s) to utilize the power of the algorithm. In the past, the algorithm had to be coded in whatever language each consumer used most often (e.g. SAS, PHP, JavaScript, etc.). Now each consumer can call one central source of truth, ensuring reproducibility and consistency.

##Installation

Two avenues for installation:

####github Because we send some data with the package, we will use pip install to get the package onto disk, as opposed to python setup.py install

git clone https://github.com/benneely/althea.git
cd ./althea
pip install --upgrade .

####pip

pip install althea

API

The goal is to have ALTHEA available via two api's:

1. python
2. RESTful

Access RESTful API resources

Once althea is installed, bring up a command line and start the RESTful server:

althea-server

By default, it will run on http://localhost:8002. Right now, there are two endpoints available to query:

1. /models - provides information on all models available
2. /models/{id}/inputs - provides information specific to a particular model

From within python we could utilize this resource as follows

import requests
import json
import pprint

#get information pertaining to all models
r = requests.get('http://localhost:8002/models')
print(r.status_code)

models = json.loads(r.text).get('result')
pprint.pprint(models)


#get information pertaining to a single model
uuid = models[0].get('model_uuid')
specific_model = requests.get('http://localhost:8002/models/'+uuid+'/inputs')
print(specific_model.status_code)
pprint.pprint(json.loads(specific_model.text))

Python API

To start playing with Althea, two risk algorithms can be loaded by default:

from althea import Metadata, Model
#add_examples is a named parameter that loads two algorithms that ship with Althea:
#Framingham 30 year CVD risk algorithm and the 2013 ACC/AHA Pooled Cohorts equation
models = Metadata(add_examples=True)
models_in_db = models.available_models()
#'deb9cc20-603f-4926-8ab9-66bb88d2ce0a'
#'cf8c7c12-0097-45ca-99cb-babcdd204d0e'

framingham = Model(models_in_db[0].get('model_uuid'))
framingham.score(sbp=120,male='No',smoke='Yes',tc=60,hdlc=110,age=33,diab='No',trtbp='No')

Add new models

To add new models to the database for query, we've create a "registrator" - a Flask application that allows users to update the sqlite database. To use, from the command line:

althea-registrator

At this time, this is only framed out, it's not functional at this time