The effect of weekly specialist palliative care teleconsultations in patients with advanced cancer –a randomized clinical trial

We conducted a two-armed, non-blinded randomized clinical trial. The study protocol (Reg 2010/382) was approved by the Committee on Research Involving Human Subjects Region Arnhem-Nijmegen and has been previously published [15]. The study was prospectively registered at The Netherlands National Trial Register (NTR2817). During the study period, two amendments were approved by the Committee on Research Involving Human Subjects. Written informed consent was obtained from all participants.

After completing the baseline measurements, the participants were randomized into two groups: intervention or control. We used block randomization with different size blocks (4 and 6) to maintain an equal balance between groups. Randomization occurred at the level of individual patients, with an allocation ratio of 1:1. The author involved in the process of approaching, informing and visiting participants (PH) was not informed about the outcome of the randomization process before baseline measurements had taken place.

Initially, it was expected that patients would be recruited by their GPs and therefore, to prevent bias, a cluster randomization procedure at the level of the GP was described in the original study protocol [15]. However, the vast majority of patients were eventually recruited via the SPCT and GPs did not recruit individual participants for this study. Consequently, there were no clusters of participants with the same GP. Therefore, in the first amendment, we decided that there was no further need for cluster randomization. As a result, randomization took place at the level of the individual participant.

Participants in both groups received palliative home care provided by their GP, supported by the SPCT according to the standard referral procedures, i.e., patients could be referred to the SPCT by their GP or by the attending hospital specialist or were not referred at all. If applicable, follow-up by the SPCT occurred by phone or by patients visiting the outpatient clinic, depending on the patient’s preference, the complexity of their problems, and/or the stage of their disease.

The primary outcome was patient-experienced symptom burden, based on the Edmonton Symptom Assessment System (ESAS) and the Hospital Anxiety and Depression Scale (HADS). The secondary and other study outcomes have been described in our study protocol [15].

Data collection ended after 12 weeks. During the 13th week of the study period a closing teleconsultation was scheduled for participants from the intervention group. During this 13th week participants did not complete any questionnaires.

Questionnaires for baseline measurements were handed over by one of the researchers (PH) or sent through postal mail. After completion, participants handed over the questionnaires to the researchers or sent them back through postal mail. During the study period, participants received and returned the required questionnaires through postal mail and sent them back every 4 weeks. If necessary, one of the researchers (PH) reminded the participants by phone, SMS, or e-mail to return the required questionnaires.

Participants completed the following questionnaires: the ESAS (at baseline, weekly follow-up), HADS, Problems and Needs in Palliative Care-short version, and a modified version of the Nijmegen Continuity Questionnaire (NCQ) (all three: at baseline, four-weekly follow-up). Informal caregivers completed one questionnaire: Self-Perceived Burden from Informal Care (EDIZ) (at baseline, two-weekly follow-up).

Additionally, participants in the intervention group completed a Patient Satisfaction Questionnaire (PSQ) after the first two teleconsultations. If applicable, the SPCT members involved in the teleconsultations and the participant’s GPs also completed a PSQ after the first two teleconsultations.

Finally, demographic information was collected at baseline. Information on other study outcomes (GP contacts, complex interventions, and hospital admissions [15]) was requested from the patient’s GP after the study period.

Data were stored and analyzed in the Radboud university medical center, Nijmegen, the Netherlands, using SPSS Software (IBM Corp. Released 2011. IBM SPSS Statistics for Windows, Version 20.0. Armonk, NY: IBM Corp).

Observed values were reported as the mean and standard deviation (SD) for continuous variables, and the number and percentage for categorical variables.

The null hypothesis for this study was that there would be no difference in patient-experienced symptom burden between the intervention and control groups. To test this hypothesis, for all outcomes, a mixed model with a random intercept for “Patient” was used to accommodate the repeated measurements over time. The dependent variable was the relevant outcome, at any time after T0. For scale variables, a linear mixed model was used, while for dichotomous outcomes, a generalized mixed model with a logit-link function was used. To identify the best model, a series of models was tested. Starting from the simplest model, each subsequent model was extended step by step until further extensions showed non-significant improvement.

The simplest model had only the Experimental condition (i.e., “group”: intervention or control) and the measurement of the outcome variable at T0 (“score at baseline”) as the independent variables. The next step added Time as an independent variable. After that, the interaction between Experimental condition and Time (“group*time”) was considered.

By using mixed models, every available observation contributes to the modeling of the relation between outcome and variables. As a result of this approach, missing data did not result in exclusion of participants from analyses.

Statistical software R, version 3.0.1, was used in combination with the lmer procedure from the lme4 library for the mixed modeling analyses.

At baseline, the mean observed TDS in the intervention group was almost 7 points higher (31.03 ± 17.21) than that of the control group (24.33 ± 14.54). Over the study period, the control group showed a slight decline in the mean observed TDS. The intervention group also showed a decline in the mean observed TDS during the first 8 weeks of the study; however, in the last 4 weeks, the mean TDS increased almost 9 points to 36.62 (±20.14) at week 12, compared to 22.38 (±11.27) in the control group (Table 2). The adjusted TDS scores became significantly higher in the intervention group, indicating a growing symptom burden in this group over time compared to the control group (Fig. 2). When testing specific points in time, this difference reached significance at week 12 (adjusted difference at week 12: 6.90; 95% CI, 0.17 to 13.63; P = 0.04; Table 3).

The mean observed HADS-anxiety scores in the control group declined somewhat during the first 4 weeks of the study (from 6.22 ± 3.91 to 5.23 ± 3.41) and thereafter remained rather stable. Within the intervention group, the mean observed anxiety scores were relatively stable during the first 8 weeks of the study, but increased between week 8 and week 12 from 7.11 (±3.39) to 8.46 (±4.25). Within the intervention group, the mean observed HADS-depression scores remained stable during the study period. The mean observed depression scores declined in the control group during the first 8 weeks (from 6.49 ± 4.57 to 5.76 ± 3.92); however, over the last 4 weeks, depression scores increased to 7.00 (±4.95) at week 12 (Table 2). The adjusted anxiety scores were significantly higher in the intervention group than in the control group (estimate effect: 1.40; 95% CI, 0.14 to 2.66; P = 0.03). The depression scores did not differ between the two groups (estimate effect: 0.30; 95% CI, −1.39 to 1.99; P = 0.73; Table 3).

The mean number of unmet needs did not differ between the intervention and control groups (estimate effect: –0.01; 95% CI, −0.07 to 0.04; P = 0.67). Additionally, the number of participants having at least one unmet need did not differ between the groups (OR: 0.79; 95% CI, 0.19 to 2.92; P = 0.66). On all three subscales of the NCQ for continuity of care, i.e., personal continuity (estimate effect: 0.15; 95% CI, −0.09 to 0.38; P = 0.22), team continuity (estimate effect: 0.16; 95% CI, −0.20 to 0.51; P = 0.39), and cross-boundary continuity (estimate effect: 0.29; 95% CI, −0.08 to 0.67; P = 0.13), there were no differences between groups. Finally, the mean number of hospital admissions during the study period did not differ between the intervention group (0.47) and the control group (0.38; P = 0.60). Study outcome measures regarding GP contacts and complex interventions did not statistically differ between both groups. The mean satisfaction scores after the first two teleconsultations were high for both the participants (90.4 ± 8.2 and 89.4 ± 9.7) and the SPCT members (87.0 ± 7.1 and 85.5 ± 15.2).

We found a difference in reported symptom burden between home-dwelling patients with advanced cancer receiving palliative care “as usual” and patients who additionally had weekly teleconsultations with a hospital-based SPCT. Therefore, we can reject our null hypothesis, which stated that there would be no difference in patient-experienced symptom burden between the intervention group and the control group. Contrary to our expectations, this additional intervention led to a higher reported symptom burden in the intervention group than in the control group. The number of unmet needs, experienced continuity of care, and reported hospital admissions did not differ between groups.

An important strength of this study is that we have systematically screened a large group of patients for participation in this trial. Despite considerable difficulties in recruitment, the intended sample size was nearly reached, although over a longer period of time and with adjusted inclusion criteria. Finally, this is the first completed RCT on telemedicine in palliative care with outcome measures that are clinically relevant and relate directly to patient care and experienced quality of life. This study also has some limitations. First, a considerable group of patients who were eligible for participation in this trial were not approached, mostly as a result of clinical considerations, which may have caused non-differential selection bias. Additionally, the relatively small number of patients who were eventually approached for participation might also reflect that offering teleconsultations in the context of a randomized study might not fit the needs of palliative care patients. Second, the attrition rate in this study was relatively high and attrition may depend on the clinical condition of participants. Although in our statistical models we have corrected for baseline measurements (i.e., the clinical condition of participants at the start of the study), study outcomes may have been influenced by participants’ worsening clinical condition during the study. Third, the participants sometimes had difficulty adequately completing the questionnaires as a result of their varying clinical conditions. This might have led to information bias, although likely non-differential. Fourth, the outcome measure “place of death” was described in the study protocol, however, it was not included in the information request at the patient’s GP; thus, this study outcome is missing. Fifth, two amendments had to be made to the study protocol to improve the recruitment rates. As a result of widening the inclusion criteria of this study, the study population may have become more heterogeneous, which may have led to a dilution of the effect of the intervention. Finally, the involvement of GPs in this study was less than expected; therefore, the participants were not recruited by their GP but were instead recruited at the outpatient clinic of the SPCT, probably leading to higher levels of specialist care in both groups, which may have positively affected symptom scores.

In this study, we introduced a model for teleconsultation in palliative care that was intended to be proactive and was mainly supply driven. Teleconsultations were scheduled on a weekly basis for a period of 3 months, irrespective of the actual needs of the patient regarding the timing and frequency of these teleconsultations. This model was shown to be ineffective in reducing experienced symptom burden, even though patients and caregivers showed a high degree of satisfaction. Therefore, we propose focusing on care models that are patient-tailored and demand-driven, i.e., patients themselves indicate when they are in need of palliative care (tele)consultations. This model could prevent a possible excess of medical care regarding palliative care, death, and dying. At the same time, to avoid a rather reactive palliative care approach, patients should be provided tools and support to guide them in proactively contacting their caregivers when problems arise.