mHealth to Early Detect Exacerbation for Older People With Heart Failure

Patient registries were carried out in two different Databases: one included general information on patients and epidemiological data. The other database stored the monitored data obtained every 48 hours.

Quality assurance and source data verification was achieved by retrieving general information on patients from the Hospital Electronic Medical Record. Moreover epidemiological information was assessed by the Principal Investigator who removed outlier values and assessed the occurrence of sharp changes in temporal series (e.g. changes >0.5Kg in two days).

Patients were recruited at the acute unit of the Geriatrics Service. When they were admitted due a heart failure exacerbation a trained nurse performed the initial assess. If the patient was suitable, a geriatrician explained to him /her the study and if the patient accepted we was asked to sign the informed consent. Then he was assigned randomly to a group. For those in the intervention group, they were trained by the geriatrician and the nurse on how to perform the measurements in the mobile phone. For those in the control group, the staff recorded the baseline characteristics, and during the follow-up the emergency visits, readmissions and death was recorded. These latter variables were also recorded in the patients of the intervention group.

The data dictionary included variables identifying patients and others describing their status and evolution:

• Oxygen saturation, measured as an integer (80-100)
• Heart rate, measured in beats/minute (0-200)
• Diastolic blood, measured in mmHg (0-100)
• Systolic blood pressure, measured in mmHg (0-100)
• Respiratory rate, measured in breaths/minute (0-100)
• Skin temperature, measured in Celsius
• Weight, measured in kg
• Glucose, measured in mg/dl (0-500)
• Short Physical Performance Battery, time to walk for 4.3 meters, measured in seconds (0-100)
• Short Physical Performance Battery, gait speed for 4.3 meters, measured in decimal meters/second (0-1)
• Short Physical Performance Battery, chair stand test once, measured in seconds (0-100)
• Symptom question Q1: Do you feel worse than the last time?, measured as a Boolean variable (YES/NO)
• Symptom question Q2: Do you feel worse than the day of the hospital discharge?, measured as a Boolean variable (YES/NO)
• Symptom question Q3: Do you feel shorten of breath?, measured as a categorical variable (effort/DLA/at rest)
• Symptom question Q4: Do you feel your legs swollen? ?, measured as a Boolean variable (YES/NO)
• Symptom question Q5: Have you woken up during the night feeling fatigue? ?, measured as a Boolean variable (YES/NO)
• Symptom question Q6: Have you needed more pillows to sleep than last time? ?, measured as a Boolean variable (YES/NO)
• Symptom question Q7: Have you taken all your drugs? ?, measured as a Boolean variable (YES/NO)
• Symptom question Q8: Have you followed the food and drinks indications of your physician?, measured as a Boolean variable (YES/NO) Sample size was calculated to reduce readmission rate by 10% in 3 months time. Having a significant level of 0.05. Sample: 50-60 per group. With the first 20 patients included, sample size was recalculated, only 40 were needed in each group. However, we included more patients in the intervention (up to 50) to face drops, only 3 dropped the study, so we obtained data from 47 in the intervention group.

Missing data were excluded from the analysis, the registry was excluded. If a patient registered more than one measurement for the same expected data, the latest remained, and the rest was deleted.

Statistical analysis:

First, we carried out a descriptive analysis of the data. Later, we analysed the effects of the intervention with telemedicine and the predictive values of the different measured variables through logistic regressions.

Descriptive data are presented as median values (IQR). Baseline characteristics between control and intervention groups were compared using non-parametric tests (Mann-Whitney).

In order to determine the effects of the intervention, the analysis was divided in two parts. First, a logistic regression model was fitted to determine if the patient monitorization (yes/no) significantly predict worsening using age and sex as covariates. Then, a new variable, hospitalization length of stay, was created with the length of stay for any readmission in each group to assess whether the use of telemonitoring affected the length of hospitalisation. Both control and intervention groups were compared using Mann-Whitney.

Finally, in the intervention group, the same logistic regression model was fitted to determine the factors that significantly predict worsening.

All analyses were made with the Statistical Package R for windows (Vienna, Austria) (http://www.r-project.org), version 3.1.1. P-value was set at level <0.05.