Deep learning for prediction of population health costs

Table 3 Shown are the top-20 codes with the highest feature importance as determined by the integrated gradients (IG) for the neural network, a high-level description of the codes as well as the corresponding IG

Code	Short description	IG
Sex Male	Male patient	0.017
GOP key 01321	Treatment by non health insurance-accredited physicians	0.014
GOP key 03004	Practitioner care of patients with age between 55 and 75 years	0.014
GOP key 03220	Treatment of a patient with at least one life-changing chronic disease	0.012
ICD-10-GM key F102	Alcohol dependence	0.011
Sex Female	Female patient	0.011
GOP key 03221	Treatment of a patient with at least one life-changing chronic disease	0.010
GOP key 03005	Practitioner care of patients older than 75 years	0.010
ICD-10-GM key F171	Mental and behavioural disorders due to use of tobacco	0.009
ICD-10-GM key M171	Unilateral primary osteoarthritis of knee	0.009
ATC code B01AC04	Clopidogrel (antithrombotic agent)	0.009
GOP key 03004R	Practitioner care of patients with age between 55 and 75 years (regional billing amendment to GOP 03004)	0.009
GOP key 32001A	Profitability bonus for arranging and providing laboratory services (regional billing amendment to GOP 32001)	0.008
GOP key 01731	Early detection of cancer in men	0.008
ICD-10-GM key M4809	Spinal stenosis	0.008
ICD-10-GM key G359	Multiple sclerosis	0.008
OPS key 80103	Application of drugs and electrolyte solutions via the vascular system in newborns	0.007
ATC code N06AX16	Venlafaxine (antidepressants)	0.007
GOP key 32001B	Profitability bonus for arranging and providing laboratory services (regional billing amendment to GOP 32001)	0.007
GOP key 07220	Surgical primary care	0.007

ISSN: 1472-6947