README.md

In md0u80c9/SSNAPStats: Stroke Sentinel National Audit Project (SSNAP) Statistics Calculations

SSNAPStats

Installation

# Install development version from GitHub
devtools::install_github("SSNAP/SSNAPStats")

Note that at present this is a private repository; to log in from the command line you should set the GITHUB_PAT environment variable. Refer to Github PAT for more information on setting this up. You can alternatively download and build the package from the site.

What is SSNAPStats?

SSNAPStats is an R package to take imported SSNAP data (currently from CSV files); break them down into cohorts, and then calculate indivdual or summary statistics from a large number of different measures upon these cohorts. These results can then be used in individual analysis, or used with the SSNAPReports and SSNAPDashboard packages for producing performance reports and real-time performance monitoring.

The data for SSNAP is hosted on the SSNAP website. This also contains the definitions for the measures.

A lot of optimisation is performed to reduce duplication and ensure that each measure is performed once and once only. This helps to produce faster results which is essential for realtime dashboarding. It also reduces the maintenance costs.

The structure of SSNAPStats could be readily applied and genericised to other national audits.

How is the package structured?

SSNAPStats divides the audit into a set of cohort definitions (that describe an audit) and a set of audit measures (that provide rules on what we are measuring). Lists of audit measures are applied to a cohort definition to produce a cohort. These functions can be applied to any audit and are quite generic.

SSNAP-specific functions sit on top of these generic functions to simplify the process by pre-defining commonly used cohorts and audit measures. The \code{ssnap_cohorts} command provides all the common cohort outputs with the least set-up. It also provides further aggregation such as \code{ssnap_scores} which is an amalgamation of various cohorts to produce the results for the scoring system used by SSNAP to reflect performance.

SSNAPStats tries to minimise the processing to the fewest steps, so we pre-parse it to work out which data will be shared between calculations so we only fetch shared information once. We also try to minimise the number of columns of data we need to only those either asked for by the cohort, or needed to calculate a value in the cohort. This is still under development and subject to more optimisation.

The cohorts code performs a series of dplyr commands upon the dataset to produce aggregated summary statistics. To maximise reusability, we use quasiquotation to separately store a list of all the possible measures for a given cohort. We then have a set of internal cohorts lists which tell us which measures are needed for a given cohort. This allows us to easily create custom cohorts with any combination of measures in, or to ensure single use when one cohort (e.g. a key indicators cohort) only wants the median of a measure, whilst another (e.g. a portfolio) wants all quartiles.

Quite often we can only calculate a summary statistic by performing calculations on each row of a field (for example time differences); we also do these in-line and store them in a list of fields. Ideally we can do these in the future at database level on record locking to improve our processing time.

Contributing to this package

Unit testing

Code should be checked in wherever possible with unit tests for proving that the given portion of code works. This packages uses the TestThat package to produce unit tests.

If you find a bug within this code, it is good practice to firstly write a unit test which proves the bug, add it to the unit tests, and then fix the bug before proving that all the other unit tests pass satisfactorily before checking it in.

Unit tests are run whenever you select Check. You can also run them independently of the full check using Cmd-Shift-T (Mac) or Ctrl-Shift-T (Windows).

Code style guide

This code is written using the Tidyverse style guide coding style. It is checked using lintr:

# Check your update meets the lintr style guide
lintr::lint_package

Check the output in the Markers tab of the Consule for any style errors and correct them. A unit test also runs lintr: so your code will fail the lintr unit test if a style error is found.

Larger code documents such as the list of ssnap measures in ssnap_measures.R are broken up using code markers that allow R studio to mark out the structure.

Documentation

Each key function (all external functions and major internal ones are documented using Roxygen2. In addition, a website for the package is created within the package using pkgdown. To update the web pages use:

# Update the package website documentation
pkgdown::build_site()

md0u80c9/SSNAPStats documentation built on Feb. 11, 2020, 11:43 a.m.