v1.0.0

Development and validation of predictive models for unplanned hospitalisation in the Basque Country: Analyzing the variability of non-deterministic algorithms

This release generates the results for the accompanying article.

The population in most developed countries is ageing, which entails growing healthcare needs. Population-wide predictive models are crucial to address this challenge and shift the healthcare system towards preventive intervention. For this reason, we predict the probability of unplanned hospitalization in the Basque Country and identify risk groups using several techniques. When dealing with non-deterministic algorithms, comparing a single model per technique is not enough to choose the best approach. Thus, we evaluate the median performance and variability of three families of models to be able to make an informed decision.

With this in mind, we achieve a double objective:

• Building predictive models for our specific application (unplanned hospitalisation in the Basque Country)
• Illustrating the need for a variability analysis in the comparison of non-deterministic algorithms

v1.1.0-alpha

First release

Prepares experiments with objective criteria that define unplanned hospitalizations:

• The events related to birth/delivery and traumatic injuries are not considered unplanned hospitalizations, even if they are administratively coded as such. This is because our project aims to identify patients at risk who are susceptible to benefit from intervention. The inpatient stays at day hospitals are also excluded. There is room for debate about all this, but we have opted for this definition of "unplanned hospitalization". In our data manipulation routines, this is coded as exclude=['nbinj','hdia'].

• We define unplanned hospitalizations following the guidelines of the CMS algorithm, instead of Osakidetza's administrative criteria, which is harder to explain formally. In our data manipulation routines, this is coded as columns='urgcms'.

Not ready

• Everything related to CCSs. Changes in generateCCS are expected.
• Everything related to explainability of neural networks
• The hyperparameter search space for neural networks will change in the future as the project develops. This release contains the space used for the manuscript.

What's Changed

• First commit to setup cluster repo by @AlexOlza in #2
• modified predict.py by @AlexOlza in #5
• Solve cluster incompatibility by @AlexOlza in #9
• Separate local development to be integrated with cluster compatible code by @AlexOlza in #10
• Attempt to fix issue #8 by @AlexOlza in #11
• Cluster by @AlexOlza in #12
• 13 config policy by @AlexOlza in #14
• Issue #8: load python 3.8 not 3.7 by @AlexOlza in #16
• Cluster by @AlexOlza in #15
• Run small random forest experiment (untitled0.py) with new config to … by @AlexOlza in #17
• 4 safe log reg dataconversion warning by @AlexOlza in #20
• modify bash scripts to avoid python downgrade by @AlexOlza in #21
• Issue25 get data negative hospitalizations by @AlexOlza in #26
• Fix random seed. Remove obsolete try/except. by @AlexOlza in #24
• Issue22 variable predictions by @AlexOlza in #29
• Train all with urgcms_excl_hdia_nbinj by @AlexOlza in #33
• Bug35 urgcms missing events by @AlexOlza in #36
• Removed obsolete SafeLogisticRegression by @AlexOlza in #39
• Add sys argv support. by @AlexOlza in #40
• Look up experiment in config file by @AlexOlza in #42
• Task41 urgcms excl nbinj by @AlexOlza in #43
• Task45 exclude tolosaldea errioxa by @AlexOlza in #46
• Task48 nested models by @AlexOlza in #50
• Doc13 config file policy by @AlexOlza in #49
• Task44 model comparison by @AlexOlza in #47
• Task48 expenditures by @AlexOlza in #51
• Calibration by @AlexOlza in #53
• Task52 additional edcs by @AlexOlza in #55
• Task52 additional edcs by @AlexOlza in #58
• Task61 hyperparameter variability by @AlexOlza in #63
• Bug64 calibration by @AlexOlza in #65
• Fix bug #66 by @AlexOlza in #67
• Enhancement70 predict compare cluster by @AlexOlza in #71
• Task68 neural networks by @AlexOlza in #69
• Pass year as environment variable by @AlexOlza in #72
• Fix PosixPath has no attribute split by @AlexOlza in #73
• Task61 hyperparameter variability by @AlexOlza in #74
• Task68 neural networks by @AlexOlza in #75
• Figures by @AlexOlza in #76
• Task68 neural networks by @AlexOlza in #77
• Figures by @AlexOlza in #78
• Task79 see by @AlexOlza in #80
• Task82 interactions by @AlexOlza in #83
• Calibrate sex comparison by @AlexOlza in #84
• Task79 see by @AlexOlza in #86
• Task79 see by @AlexOlza in #88
• Task68 neural networks by @AlexOlza in #81
• Model comparison figures by @AlexOlza in #90
• Pass past datasets to calibration by @AlexOlza in #92
• Model comparison figures by @AlexOlza in #94
• Task93 no case mix by @AlexOlza in #95
• Task93 no case mix by @AlexOlza in #96
• Add patient counter by @AlexOlza in #97
• Fix typo in filename by @AlexOlza in #99
• Fix issue #100 by @AlexOlza in #101
• Task93 no case mix by @AlexOlza in #103
• Task104 figures tables by @AlexOlza in #105
• Task93 no case mix by @AlexOlza in #107
• Check diabetes by @AlexOlza in #112
• Task104 figures tables by @AlexOlza in #110
• Minor fixes by @AlexOlza in #115
• Task104 figures tables by @AlexOlza in #119
• Fix bug #116 CCS needs age and sex by @AlexOlza in #120
• Task104 figures tables by @AlexOlza in #122
• Add comorbidities to descriptive table by @AlexOlza in #123
• Task93 no case mix by @AlexOlza in #121
• Bug125 by @AlexOlza in #126
• Fix 127 by @AlexOlza in #128

Full Changelog: https://github.com/AlexOlza/estratificacion/commits/v1.0.0-alpha