The ACM is a software conceived as a tool to help the clinician in making decisions on the prescription of ESA and iron for the treatment of anaemia in haemodialysis patients. It is a model of artificial neural network that uses updated individual data to predicts Hb values based on a prescribing algorithm of darbepoetin and iron. ACM integrates data through an automatic interface module integrated into the clinical data system of FMC, EuCliD®. The integrated data includes anthropometric as sex and height that extracts from the moment of admission, the latest analytical data including Hb, ferritin, saturation index of transferrin (IST), calcium, phosphorus, sodium, potassium, leukocytes, C-reactive protein (CRP), mean corpuscular volume (MCV), mean corpuscular haemoglobin (MCH) and albumin, data from the last dialysis session as dry weight, predialysis weight and Kt/V measured by online clearance monitoring (OCM) and doses of darbepoetin and iron used in the last 90 days, as well as the Hb change with respect to the previous month.

Each time that a Hb value is introduced in the ACM, the program generates a prescribing suggestion of darbepoetin and iron sucrose (Venofer®) in the case of being accompanied of iron kinetic data. This suggestion is a recommendation that always requires the validation of the clinician. If the suggestion is rejected, the clinician must formulate a different prescription and indicate the cause of the non-confirmation (Appendix A). ACM considers temporarily non-eligible to apply the model the patients in the following situations: under 18 years, admission in less than 90 days, transfusion in the last 90 days or insufficient number of dialysis sessions (at least 27 of the 39 expected), either by hospitalisation or vacation.

ACM was launched at the dialysis clinics of FMC Services Murcia in Cartagena and San Pedro del Pinatar in June 2014. The use of ACM was subsequently discontinued during 6 months (November 2014 to April 2015) so it was resumed in May 2015. This interruption period is considered the control phase; therefore, the study was carried out during a total of 18 months (May 2014 to October 2015). The first period (May 2014 to October 2014) is considered intervention phase 1, during which nephrologists use ACM support for the treatment of anaemia. The second period (November 2014 to April 2015) is considered the control phase, in which the treatment of anaemia is carried out in a traditional manner without the ACM support, using the trial-error method to obtain values of Hb between 10 and 12g/dl, according to the protocol of the clinics based on the European Best Practice Guidelines for Anaemia in patients with chronic renal failure.41

During the third period (May 2015 to October 2015), or intervention phase 2, nephrologists again use ACM support for the management of anaemia.

During the study, in both clinics the plan pre-established for analytical monitoring was maintained, Hb levels were measured monthly and iron kinetics (ferritin and IST) were obtained bi-monthly. Controls of the water quality were performed according to the FMC protocol; the results were similar in the different phases of the study complying with the guidelines of the of the Spanish Society of Nephrology.42 Likewise, both in the Cartagena Clinic and in San Pedro del Pinatar, ultrapure water is available to perform hemodiafiltration (HDF) treatments online.

The effects of the use of ACM were evaluated both in the haemodialysis unit and in each individual patient.

All patients included were older than 18 years with at least one haemodialysis session and one Hb determination during the control or intervention phase. Thus, there were 213 patients in the intervention phase 1, 219 in the control period and 218 in the intervention phase 2. The primary outcome was the percentage of Hb values in range and the median dose of EEA administered (expressed as dose/patient/kg/month or dose/patient/month). The Hb level was considered in range if values were between 10 and 12g/dl, or >12g/dl in the absence of treatment with EEA (i.e., without EEA from at least 35 days before the Hb measurement).

The results were calculated for the entire population as a measure of the global impact of ACM. In addition, we performed a subanalysis focused only on the confirmed ACM suggestions, in this case the Hb measurements and the doses of drugs recommended by ACM were separated.

The secondary outcome included the need for transfusion and hospitalisation, the occurrence of cardiovascular events or death. Cardiovascular events were defined as the presence of any of the following: death or appearance of cardiac or cerebrovascular pathology or admissions due to this cause. These events were obtained from the clinical history identified by an ICD-10 code contained in the I00–I99 interval, except codes I80–I89 relating to diseases of the veins, vessels and lymph nodes.

The analysis of the individual patient required at least 5 measurements of Hb in the same dialysis centre during a given period. It was necessary to have a significant number of Hb measurements in each patient to evaluate the fluctuation of them over time. The number of patients fulfilling this requirement were 173 patients in the intervention phase-1, 184 patients in the control phase and 188 patients in the intervention phase-2. The individual variability of Hb was estimated by the standard deviation (SD) of Hb in the three phases.

To assess the ACM contribution, we considered the percentage of Hb values in range and the consumption of EEE discriminating between the ACM suggestions that were confirmed, rejected or null.

The same parameters were also evaluated in the subpopulation of patients with the majority of suggestions accepted (2/3 or approximately 66% of suggestions accepted. The number of patients in the ACM compliant group were 111 the intervention phase-1 and 154 in the intervention phase-2.

Finally, the primary results were analyzed separately in groups of patients according to vascular access (native arteriovenous fistula, prosthetic vascular access or catheter).

The statistical analysis was performed in Matlab®. T-test was used to compare variables with normal distribution and the Wilcoxon test for variables with non-normal distribution. Fisher's test was used to compare proportions. The data are expressed as mean±standard deviation (SD) or median and interquartile range (ICR) depending on the variable. For all tests, a p<0.05 value was considered statistically significant.

Baseline characteristics of haemodialysis patients participating in the study are shown in Table 1.

The primary outcome (percentage of Hb in range and consumption of EEA) and the secondary results (deaths, hospitalisations, cardiovascular events and transfusions) are presented in Table 2. During the control phase, the average monthly consumption of darbepoetin increased a 33% in the entire population (from 20μg/month [70] in the intervention phase-1 to 30μg/month [80], p=0.032) and was reduced again in the intervention phase-2 but without reaching statistical significance (20μg/month [80] p=NS). At the same time the percentage of Hb values in range fell by 5.5% (from 78.2 to 72.7%, p=0.002) and then increased by 8.2% (from 72.7 to 80.9%, p<0.001) with the reintroduction of ACM.

The analysis of results including only accepted suggestions, both Hg in range and dose of EEA showed a decisive improvement (85.4% of Hb values in range with a median use of darbepoetin of 10μg/month [40] in the intervention phase-1 and 86.2% of Hb values in range with a median of darbepoetin=20μg/month [60] in the intervention phase-2.

The percentage of Hb measurements above the range was higher during the control phase than the intervention phases (intervention phase-1: 8.24%; control phase: 13.55% (p<0.001); intervention phase-2: 7.42%, p<0.001); this effect was even more remarkable if only the laboratory results obtained from the accepted ACM suggestions were considered.

The monthly iron consumption per patient also decreased during the two intervention periods, but this decrease was only significant in the intervention phase-2 (intervention phase-1: 100mg/m [200], control phase: 150mg/m [300], p=NS, intervention phase-2: 100mg/m [200], p<0.001).

Other relevant analytical differences were also observed. It should be noted that during the intervention phases, patients had reduced levels of CRP (intervention phase-2: 4.66mmol/l, control phase: 6.68mmol/l, p=0.009) and of ferritin (intervention phase-1: 459ng/ml; control phase: 511ng/ml, p=0.005) and higher levels of albumin (intervention phase-1: 3.93g/dl, control phase: 3.84g/dl, p=0.001). In addition, during the intervention phase-1, a greater number of HDF treatments were performed, and no significant differences in the rest of the study periods (intervention phase-1: 66.2%, p=0.004; control: 52.51%; intervention phase-2: 51.38%, p=NS).

The effect of vascular access on the primary outcome was analysed separately. The results are presented in Table 3 using the same indicators but divided into two categories: native or prosthetic arteriovenous fistula (AVF) versus catheter. It is striking to observe that only in the group of patients with AVF the consumption of darbepoetin varies throughout of the different stages of the study; thus, it increases by 34% during the control phase and decreases by 16% after resuming the use of MCA during the intervention phase-2. In the group of patients with a catheter, the consumption of darbepoetin did not change after the use of ACM and was also 30% higher as compared to the consumption in the group with AVF.

Concerning the secondary results, the incidence of cardiovascular events had a sustained decrease over time that did not seem to be dependent on the use of ACM (intervention phase-1: 847.50/1000 patients-year, phase control: 626.09/1000 patients-year, p=0.009, intervention phase-2: 441.23/1000 patients-year, p=0.002). Death events seem to be minor during the intervention phases. It is only significant if the intervention phase with the control phase are compared (intervention phase-1: 1.45%, control phase, 5.48) %, p=0.032, intervention phase-2: 2.75%, p=NS). The number of patients requiring transfusions was similar in the three periods of the study.

The characteristics of the cohort of patients considered for the analysis are presented in Table 4. These are all patients selected from the intervention phase-1, control phase and intervention phase-2, and patients that complied with ACM in intervention phases 1 and 2. With the use of ACM in the intervention phase-2 there was a significant reduction in Hb fluctuation (from 0.91±0.49 to 0.82±0.37g/dl; p=0.04), and a significant increase in the percentage of patients with at least 2/3 of their Hb values in range (from 70.1 to 84.0%, p=0.001).

These results are confirmed and even accentuated in the ACM compliant group. For these patients, the fluctuation of Hb was further reduced in the intervention phases, but it was significant in the intervention phase-2 (0.83±0.41g/dl in the intervention phase-1, p=NS, and 0.77±0.35g/dl in the intervention phase-2, p<0.001) and it was also observed an increase in the percentage of patients with the majority of Hb values in range (84.7% in the intervention phase-1, p=0.005, 90.3% in the intervention phase-2, p<0.001).

Secondary outcomes also tended to be reduced during the use of ACM, particularly in the ACM compliant group, where a significant decrease in the incidence of transfusions and days of hospitalisation per 1000 patient-years were observed (although this is not confirmed in the global analysis). Conversely, cardiovascular events show a downward trend over time that does not seem to be dependent on the use of ACM.

The primary results obtained according to the type of vascular access in the individual patient are presented in Table 5. The data are divided into two categories: native/prosthetic AVF versus catheter. During the intervention phase-2, the group of patients with AVF have a decrease in Hb variability (from 0.88±0.44 to 0.77±0.35, p=0.02), as well as an increase in the percentage of patients with Hb in range (71.85–86.67%, p=0.002). This is more remarkable in the group of confirmed suggestions (ACM compliant group), where the variability of Hb is further reduced, reaching 0.74±0.34 in the intervention phase-2, and the percentage of patients with >66.6% Hb in the range exceeded 90% during the two intervention phases. The consumption of darbepoetin was increased during the control phase, while the percentage of patients with Hb in range was reduced. The group of patients with catheter have a greater consumption of darbepoetin as compared to the group of AVF, but without significant changes in the different phases of the study.

We should state that the algorithm has been designed in such a way that it incorporates data on the pharmacokinetics and pharmacodynamics of darbepoetin with a special focus on the erythrocyte maturation time and its lower survival in situations of renal failure. It also uses anthropometric, laboratory, dialysis dose and clinical data to come out with the dose suggestion. On the other hand, MCA considers that the current Hb levels are influenced by the doses of darbepoetin received in the last 3½ months and it bases the recommendations on the doses administered (not those prescribed) taking into account the administration schedule.50 All this together makes the ACM to achieve a great precision, minimizing the fluctuations of human prescription.

Looking at the number of prescription confirmations throughout the study, we observed that number of suggestions accepted increased from intervention phase 1 to intervention phase 2. This translates into a distrust by the nephrologist on the suggestions proposed by the new tool, rejecting a large number of suggestions and obtaining worse results, while in the second phase of intervention, the safety of the model seems to increases, which translates into a greater number of accepted suggestions resulting in more patients with Hb in range and a lower variability in Hg levels. The growing percentage of acceptance makes us think that nephrologists not only agree in simple cases clinically stable, but the is a high confidence in the program, which is demonstrated in the improvement of results at a general level.

During the last decade there have been different prediction algorithms used to improve the treatment of renal anaemia and individualize the dose of ESA in dialysis patients; and the results have been promising results. Noteworthy is the prospective and multicenter study published by McCarthy et al.,51 which describes improvements in Hb variability and a 40% reduction in the dose of darbepoetin in a large population of haemodialysis patients when a biomedical system is applied in patients treated with iron. Although the results seem similar to ours, there are notable differences in both studies that makes comparisons difficult, such as the use of higher doses of darbepoetin both at the beginning and at the end of the study and a greater range of target Hb (10–12.9g/dl).

One of the limitations of the present study is that, despite of being prospective, the value of n is not very large and the follow-up time is insufficient to assess the impact of ACM on cardiovascular morbidity and mortality. Given the characteristics of the program and the need to have sufficient detailed and updated information for each patient, its validation in incident patients may be difficult to analyse (given the small number of incident patients); hospitalisations and vacations results in loss of follow-up and subtracting data to the program. We should consider that it is likely that an analysis including prevalent patients without hospitalisations and with AVF would have reflected more accurately the capabilities of ACM by itself. However, we have tried to reproduce the usual clinical practice while maintaining all factors that the nephrologist has to deal with. Another limitation to consider is that the program has only been validated with darbepoetin and in the haemodialysis population, thus, it cannot be used in other clinical scenarios.

In conclusion, in our study, MCA was an effective tool to help the clinician, improving the results of anaemia in patients on haemodialysis, minimizing the risks of treatment with ESA and reducing costs.