Artificial Intelligence for Optimal Anemia Management in End-stage Renal Disease: The Anemia Control Model (ACM) Trial

Fresenius Medical Care has developed an algorithm that uses a data-driven computational intelligence model based on an artificial neural network architecture (ACM) to generate individualised ESA dose recommendations from a history of dose and response information and specific patient demographic characteristics.

The ACM has been validated and complies with the European requirements for medical devices. The ACM was classified as a Class I medical device in accordance with Directive 93/42/EEC. A proof of concept trial was conducted in 3 NephroCare dialysis clinics (managed by Fresenius) in the Czech Republic, Portugal, and Spain. It could be shown that the introduction of ACM-guided therapy led to a significant decrease in median darbepoetin doses and to a significant increase in on-target haemoglobin (Hb) values along with a decrease in Hb fluctuation. Moreover, a retrospective trial conducted in NephroCare clinics in Portugal, suggested that ACM is able to reliably predict the long-term response to ESA and iron therapy in patients undergoing haemodialysis.

The current trial will test the applicability of the ACM outside of Fresenius clinics in a public hospital setting in the UK. Both intravenous (IV) iron and ESA doses will be recommended by the algorithm. The effectiveness of ACM-guided therapy on several anaemia outcomes will be assessed in adult patients with End Stage Renal Disease (ESRD).This trial will be conducted at 1 main unit and 5 satellite units at King's College Hospital, London, UK, in patients with ESRD who are routinely undergoing haemodialysis.

This trial is designed to assess the effectiveness of the ACM software on anaemia management in routine clinical practice. However, all ultimate decisions on therapeutic or diagnostic procedures, treatments, management of the disease, or resource utilisation will be at the discretion of the Investigator.

The trial will consist of a retrospective control period and a prospective period. During the prospective period, the ACM will be used to facilitate the Investigators' decision making and will help the Investigators to administer a personalised IV iron and ESA therapy, whereas treatment according to standard of care will be documented retrospectively for the same patients during the retrospective period of the trial. Thus, patients can serve as their own control.

The planned overall duration of the trial is 18 months (12 months recruitment period + 6 months until last patient last visit). The planned duration of prospective treatment for an individual patient will be 6 months. The collection of retrospective data from medical records covering a period of 6 months will take place as soon as the Informed Consent Form (ICF) is signed but at the latest at baseline.

Data will be collected at designated time Points (monthly) throughout the Trial once the ICF is signed: at the latest at baseline (collection of retrospective data), at baseline (start of prospective documentation), and for the observation time points (Month 1 to Month 6). However, examinations will follow routine clinical practice at the site according to the Investigator's decision.

• Anemia Control Model (ACM) (Device): The ACM is mainly composed of 2 sub-Systems - predictor model which, depending on the input data, forecasts the response to anaemia drug therapy for a specific patient. The predictor model is implemented as a feed-forward artificial neural network. The ACM is an algorithm that extracts the optimal policy to achieve the established clinical outcome for anaemia management using the predictor model.
• IV iron (Drug), also known as Venofer, iron sucrose: IV iron given monthly as required - dose determined by the ACM and as agreed by the investigator
• Erythropoiesis-Stimulating Agent (ESA) (Drug), also known as epoetin beta: ESA given monthly as required over 6 months - dose determined by the ACM and as agreed by the investigator

Primary outcomes

• Change in the proportion of patients with haemoglobin within the target range as compared to the historical control period (non-inferiority testing) [Month -6 to Month -1 compared with Month +1 to Month +6]: The proportion of patients with at least 5 (standard of care, approximately monthly) Hb measurements and with 80% of these measurements within the target range of 10 to 12 g/dl from Month -6 to Month -1 will be compared with the proportion of patients with at least 5 measurements and with 80% of these measurements within target range of 10 to 12 g/dl from Month 1 to Month 6 (non-inferiority testing).

Secondary outcomes

• Change in the proportion of patients with haemoglobin within the target range as compared to historical control period (superiority testing) [Month -6 to Month -1 compared with Month +1 to Month +6]: The proportion of patients with at least 5 (standard of care, approximately monthly) Hb measurements and with 80% of these measurements within the Hb target range of 10 to 12 g/dl from Month -6 to Month -1 will be compared with the proportion of patients with at least 5 measurements and with 80% of these measurements within target range of 10 to 12 g/dl from Month 1 to Month 6 (superiority testing).
• Change in haemoglobin fluctuations as compared to historical control period [Month -6 to Month -1 compared with Month +1 to Month +6]: The Hb fluctuation in patients with at least 5 (standard of care, approximately monthly) Hb measurements from Month -6 to Month -1 and from Month 1 to Month 6 as measured by Hb, Coefficient of Variation (CV), and Standard Deviation (SD) from Month -6 to Month -1 period versus Month 1 to Month 6.
• Change in cumulative ESA dose as compared to historical control period [Month -6 to Month -1 compared with Month +1 to Month +6]: Cumulative ESA dose from Month -6 to Month -1 versus cumulative ESA dose from Month 1 to Month 6.
• Change in cumulative IV iron dose as compared to historical control period [Month -6 to Month -1 compared with Month +1 to Month +6]: Cumulative IV iron dose from Month -6 to Month -1 versus cumulative IV iron dose from Month +1 to Month +6.

Sex: All, ages: 19 Years to 90 Years

Inclusion Criteria:

* Age 19 to 90 years
* On haemodialysis for the past 18 months prior to baseline
* Treatment with IV iron sucrose during the past 6 months according to the respective Summary of Product Characteristics (SmPC)
* Treatment with epoetin beta during the past 6 months according to the respective SmPC
* Regular Hb measurements and at least 5 (standard of care, approximately monthly) Hb measurements during the past 6 months
* Ferritin measurements during the past 6 months (at least 2 measurements)
* Signed informed consent

Exclusion Criteria:

* Life expectancy <6 months
* One or more Hb measurements <8 g/dl during the control period
* Living-donor transplant scheduled within the next 6 months
* Scheduled for switch to peritoneal dialysis or home haemodialysis
* Blood transfusion during the past 9 months
* Pregnancy or breast feeding
* Active infection
* Current malignancy or haematological disorder
* Previous severe hypersensitivity reaction to IV iron sucrose
* Serious allergic reactions to darbepoetin alfa or epoetin alfa/beta/zeta, respectively
* Current treatment with PEGylated erythropoietin
* Surgery in the past 6 months
* Surgery scheduled within the next 6 months
* Participation in a clinical trial in the past 7 months

• Barbieri C, Mari F, Stopper A, Gatti E, Escandell-Montero P, Martinez-Martinez JM, Martin-Guerrero JD. A new machine learning approach for predicting the response to anemia treatment in a large cohort of End Stage Renal Disease patients undergoing dialysis. Comput Biol Med. 2015 Jun;61:56-61. doi: 10.1016/j.compbiomed.2015.03.019. Epub 2015 Mar 23. PMID 25864164
• Barbieri C, Molina M, Ponce P, Tothova M, Cattinelli I, Ion Titapiccolo J, Mari F, Amato C, Leipold F, Wehmeyer W, Stuard S, Stopper A, Canaud B. An international observational study suggests that artificial intelligence for clinical decision support optimizes anemia management in hemodialysis patients. Kidney Int. 2016 Aug;90(2):422-429. doi: 10.1016/j.kint.2016.03.036. Epub 2016 Jun 2. PMID 27262365
• Barbieri C, Bolzoni E, Mari F, Cattinelli I, Bellocchio F, Martin JD, Amato C, Stopper A, Gatti E, Macdougall IC, Stuard S, Canaud B. Performance of a Predictive Model for Long-Term Hemoglobin Response to Darbepoetin and Iron Administration in a Large Cohort of Hemodialysis Patients. PLoS One. 2016 Mar 3;11(3):e0148938. doi: 10.1371/journal.pone.0148938. eCollection 2016. PMID 26939055

• Iain Macdougall, Principal investigator, King's College Hospital NHS Trust