Would Artificial Neural Networks Implemented in Clinical Wards Help Nephrologists in Predicting Epoetin Responsiveness?

Luca Gabutti; Nathalie Lötscher; Josephine Bianda; Claudio Marone; Giorgio Mombelli; Michel Burnier

Disclosures

BMC Nephrology 

In This Article

Background

Stable haemoglobin levels maintained in the target range of 11 to 12 g/dL as recommended by the Kidney Disease Outcomes Quality Initiative, are associated with both clinical and quality of life benefits as well as a reduction in hospitalisation and mortality.[1,2,3,4] Whether, in targeted subgroups, the haemoglobin concentration should be set above 12 g/dL has not been definitively demonstrated[5]; however, considering the likelihood of increasing thrombotic events,[6,7] its value should not exceed 14 g/dL.[5]

The response to erythropoietin is known to have a large inter- and intra-individual variability explained by blood losses, co-morbidities,[8] dialysis efficiency,[9,10] iron status,[5,11] folic acid and vitamin B12 deficiency,[12] hyper- and hypo-parathyroidism,[13,14] pro-inflammatory cytokine activities,[15] aluminium toxicity,[14] treatment with angiotensin converting enzyme inhibitors (ACE-I)[16] and probably angiotensin II receptor blockers (ARB).[17] Thus, maintaining the haemoglobin level in the target range is sometimes a difficult task which necessitates regular doses adjustments.

To optimize anaemia management several protocols, based on physician or nurse-driven algorithm as well as computer assisted prescription tools, some of them involving the use of Artificial Neural Networks (ANNs), have been described.[18,19,20,21,22,23,24,25,26]

A large amount of clinical and biochemical data that could be useful in making crucial follow-up decisions are actually collected during dialysis sessions.[27,28,29] Unfortunately the multidimensionality and at least partial non-linearity of the data, (i) limits the value of both intuition/experience of the nephrologists and standard statistical procedures and (ii) makes their interpretation and practical use in clinical wards difficult.[27] The importance of individualizing drug dosage regimens by adding patient-specific post-administration data about serum levels or responsiveness to population pharmacokinetic and dynamic models, has been thoroughly demonstrated.[30] Compared to other non linear mathematical and statistical tools based for instance on Bayesian fitting and adaptive control, ANNs have the advantage of being user friendly, tolerating missing data and errors in individual variables well and also of being applicable to translate multivariate non-linear relationships into continuous functions without the need of understanding precisely the underlying relationships between variables.[31,32,33,34,35,36] ANNs have been widely used in clinical medicine and have already assisted nephrologists in solving various complex clinical problems.[27,28,29,30,31,32,33,37,38]

The purposes of the present study were (i) to characterize the linear or non-linear relationships between several clinical and biological variables and the response to epoetin beta and (ii) to build a computer assisted mathematical tool able to predict the epoetin requirement in an individual patient and the monthly adjustments in the epoetin dose.

Comments

3090D553-9492-4563-8681-AD288FA52ACE
Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.

processing....