A predictive index of intra-dialysis IDH. A statistical clinical data mining approach.

Intra-Dialysis Hypotension (IDH) is one of the main hemodialysis related complications, occurring in 25-30% of the sessions. The factors involved in the onset of hypotension in patients undergoing dialysis are due both to clinical conditions (e.g. presence of vascular or cardiac diseases, neuropathology, anemia) and treatment settings such as temperature of the dialysate, sodium concentration, buffer composition, ultrafiltration rate, etc. The patient’s peculiar reaction to the treatment implies difficulties in preventing IDH episodes. This work explores the possibility to use a multivariate analysis of clinical data to quantify the risk to develop IDH at the beginning of each session. The study is framed in the DialysIS project (Dialysis therapy between Italy and Switzerland) funded by INTERREG – Italy – Switzerland and Co-funded by European Union. Data referring to a total of 516 sessions performed on 70 adult patients undergoing dialysis treatment (50 patients enrolled at A. Manzoni Hospital Lecco, Italy and 20 patients at Regional Hospital of Lugano, Switzerland) were collected. Clinical prescriptions, hydration status, dialysis machine data and hematochemical data were recorded and stored in a unique flexible structured MySQL® database. A statistical analysis was performed to find the potential risk factor related to IDH onset. IDH episodes were automatically detected during the monitored sessions, according to the literature criteria. Patients suffering from IDH in 2 or more sessions were classified as Hypotension Prone (HP), the others as Hypotension Resistant (HR). Initial values of potassium concentration [K+], systolic (SBP) and diastolic (DBP) blood pressure, and weight gain (ΔW) from the end of the previous treatment result to be statistically different between the HP and HR groups. A new index, J, was defined as a weighted patient-specific combination of these parameters and calculated for each session of each patient. The weight of the index coefficients can be dynamically adjourned based on the longitudinal analysis of [K+], SBP, DBP, and ΔW. The results reported in this paper were calculated based on a longitudinal analysis of a minimum of three sessions for each patient. The accuracy of the J index in predicting IDH events has been evaluated and quantified in terms of percentage number of predicted IDH events, with respect to the total number of IDHs. Values of J index higher than 1 point out the risk of IDH onset. J allows the prediction of 100% of IDH episodes using 5 sessions, the 90% using 3 sessions. More specifically, at Lecco Hospital 43 IDH events were detected by the automatic system of which 100% and 95% were respectively predicted by the new index calculated using 5 or 3 sessions. Similarly, at Lugano Hospital 58 IDH were detected by the automatic system of which 100% and 87,5% were predicted using 5 or 3 sessions respectively. A longer longitudinal dataset will allow a higher matching of J to actual IDH episodes. In conclusion, the evaluation of this new index at the beginning of the dialysis session prior to connecting the patient to the machine can provide the clinician with useful information about the risk for the patient to develop cardiovascular instabilities (IDH) during the treatment and can advise the physician about the need to modify the prescription.