shukra

Tools for meta-analytical statistics, including Network Meta-Analysis (NMA), in NodeJS

Example

A fixed effects NMA on difference in means:

const { meanDifferenceNMA } = require('shukra/nma');

const studies = [1, 1, 2, 2, 2, 3, 3];
const trts = ['A', 'B', 'A', 'B', 'C', 'C', 'B'];
const means = [8, 10, 7, 10.5, 10.5, 10, 11];
const sds = [4.92, 3.87, 3.25, 6.35, 6.66, 4.32, 4.30];
const ns = [63, 45, 35, 44, 53, 75, 29];

const nma = meanDifferenceNMA(studies, trts, means, sds, ns, false);

nma.getEffect('A', 'B');
/*
-2.819
 */

nma.computeInferentialStatistics('A', 'B');
/*
{
  p: 0.000008,
  lowerBound: -4.06,
  upperBound: -1.58
}
*/   

Computing a pooled mean under random effects using inverse variance weighting:

const { pooledMean } = require('shukra/pooling');
pooledMean(ns, means, sds);
/*
{
  estimate: 9.493324459831722,
  lower: 8.334404617356348,
  upper: 10.652244302307096,
  studyEstimates: [
    { lower: 6.785093322564656, estimate: 8, upper: 9.214906677435344 },
    { lower: 8.86928592265621, estimate: 10, upper: 11.13071407734379 },
    { lower: 5.923293264578356, estimate: 7, upper: 8.076706735421643 },
    ...
  ]
}
*/

Install

Install the current release on npm:

npm install shukra

Install the most recent development version:

npm install --save https://github.com/holub008/shukra/tarball/master

About Shukra

shukra is a web targeted meta-analytical statistics toolkit. It was specifically designed for simplicity & speed over comprehensiveness (as in formal research publication); this means shukra can fit an NMA in the scope of a web request (<10 milliseconds).

shukra is developed for and by Nested Knowledge, but we welcome all users and contributions. It is predominantly transcribed from the meta and netmeta packages, and carries their licenses. These packages are also used for verifying correctness (see test/ for code samples).

The feature set is growing, but currently offers several modules:

• NMA (shukra/nma)
  • Random & Fixed effects models
  • Mean Difference (continuous outcomes) & Odds Ratios (binomial outcomes)
  • Inferential statistics on effect size estimates and individual study point estimates
• Pooling (shukra/pooling)
  • Inverse variance weighting
  • Random & Fixed effects models
  • Imputation of missing data
  • Inferential statistics on pooled estimates and individual study point estimates

About NMA

Network Meta-Analysis is a model for performing research synthesis. Given multi-arm (2+) studies, where each arm represents a discrete experimental treatment, and a measured outcome for each arm, an NMA produces an effect size estimate and associated inferential statistics.

NMA is exciting in that it models direct contrasts (i.e. head to head treatment comparisons in studies) and indirect evidence (i.e. using knowledge about A vs. B in one study and B vs. C in another to make inference on A vs. C).