Remember me
Managing symptoms during cancer treatment is essential for patients’ quality of life, workability, and performance []. Symptoms such as pain, fatigue, and gastrointestinal problems commonly lead to emergency department visits []. Emergency department visit rates appear to be higher among patients with cancer than in the general population, although the magnitude or underlying reasons for this remain understudied [].
Electronic patient-reported outcome (ePRO) interventions have gained recognition as convenient and safe tools for promoting the early detection of symptoms and adverse events [,]. Collecting ePROs has demonstrated high acceptance [-], long-term feasibility [], and positive outcomes related to physical and psychological symptoms [-], as well as increased survival []. ePROs are also suggested to help mitigate unplanned acute care and unnecessary hospitalizations during cancer treatment; however, this assertion requires more robust empirical confirmation [,]. In our studies, the use of the interactive app Interaktor was associated with a decreased symptom burden during radiotherapy (RT) for prostate cancer [], neoadjuvant chemotherapy (NACT) for breast cancer [], and up to 6 months after surgery for pancreatic cancer [].
Health economic evaluations are essential for supporting the implementation of cost-effective interventions [,] and guiding decision-makers []. A cost-utility analysis (CUA) is one type of health economic evaluation that compares the costs and health outcomes of medical treatments or care by estimating the cost required to improve a unit of health outcome []. Quality-adjusted life years (QALYs) is a generic measure of disease burden that accounts for both life quality and quantity. One QALY corresponds to 1 year of perfect health, while 0 represents death []. In Sweden, the National Board of Health and Welfare (NBHW) has defined a cost per QALY of €9685 (€1=US $1.03) as low, more than €48,423 as high, and more than €96,846 as very high [].
Most health economic evaluations of ePRO interventions have focused on patients with advanced or metastatic cancer []. Lizée et al [] demonstrated the cost-effectiveness of ePRO from a national health insurance perspective, despite increased costs, due to associated survival benefits. Velikova et al [] evaluated the cost-effectiveness of ePRO for patients undergoing systemic treatment for colorectal, breast, or gynecological cancer, comparing the cost per additional QALY gained at 18 weeks after randomization from both health care and societal perspectives. The analysis considered costs for the intervention manual, software maintenance, and patient time off work but excluded intervention development costs. No significant cost differences were observed between the intervention and usual care groups. The study indicated a 55% likelihood of cost-effectiveness at the National Institute for Health and Care Excellence cost-per-QALY threshold.
This study was conducted alongside 2 randomized controlled trials (RCTs) of the ePRO intervention Interaktor during NACT for breast cancer (B-RCT) and RT for prostate cancer (P-RCT). The primary aim is to evaluate the cost utility of the app for ePRO and interactive support from the health care provider’s perspective (Region Stockholm Health Care Organization). Additionally, the study examines the impact on patients’ health care utilization and associated costs.
The research was approved by the Swedish Ethical Review Authority (permit numbers 2013/1652-31/2 and 2017/2519-32). Written informed consent was obtained from all patients at the time of study inclusion. Data were deidentified to protect participants’ privacy. Patients received written and verbal information about their right to opt out without affecting their subsequent care. No compensation or payment was provided for participation.
Study DesignBetween 2016 and 2019, Interaktor was evaluated through 2 parallel prospective open-label RCTs, with symptom burden as the primary endpoint, measured using the EORTC QLQ-C30 version 3.0 []. Patients were consecutively recruited from oncology clinics at 2 university hospitals in Stockholm, Sweden. Eligible and interested patients met with a researcher who provided detailed information about the trial. Refer to the previously published study protocol and clinical effectiveness article [,] for details on the eligibility criteria, intervention design, and randomization process. No changes were made to the methods after the protocol was registered (NCT02479607 and NCT02477137) and the trials commenced.
SamplesOne RCT included a sample of patients with breast cancer treated with NACT (B-RCT), and the other included a sample of patients with prostate cancer treated with RT (P-RCT). In both RCTs, patients were randomly allocated to the intervention or control group. In the B-RCT, there were 74 patients in the intervention group and 75 patients in the control group. Of these, 69 (93.2%) in the intervention group and 71 (94.7%) in the control group completed the follow-up and were considered complete cases ( and ). In the P-RCT, a total of 150 patients were randomly allocated to the intervention (n=75) or control (n=75) group. Of these, 58 (77%) in the intervention group and 56 (75%) in the control group completed the follow-up questionnaires and were considered complete cases [] (-4). The sample size for both RCTs was estimated based on an effect study conducted with patients receiving RT for prostate cancer [], with symptom distress as the primary outcome. The effect size difference (Cohen d=0.54) indicated that, for 90% power at P<.05, 71 patients were required in each group.
Intervention and Standard CareThe Interaktor smartphone and tablet app is an ePRO intervention designed for daily symptom reporting and interactive support during cancer treatment. It includes a symptom questionnaire, graphs of symptom reporting history, self-care advice related to disease and treatment-associated symptoms, and links to websites with additional information. Oncology ward nurses are alerted via SMS text messages when severe symptom levels are reported. Nurses can access patients’ reports through a web interface, which facilitates patient-clinician communication. Depending on the alert, nurses contact the patient within 1 hour or 1 day. The Interaktor versions used in this study did not include any institutional affiliation display or logo.
Patients in the intervention group reported daily via the Interaktor app on weekdays, starting from their first day of treatment and continuing until 2 weeks after treatment in the B-RCT (mean treatment duration: 15 weeks in both groups) and, until 3 weeks after treatment in the P-RCT (mean treatment duration: 5 weeks in both groups). In the intervention groups, registered nurses at the patients’ oncology units responded to the symptom report alerts. Additionally, a researcher was available to assist with any technical questions or issues. Outside office hours, patients were advised to contact health care personnel according to the standard procedure of their oncology clinic. The intervention and app content remained unchanged during the evaluation process. Patients received daily reminders if a report had not been submitted. A comprehensive description, including screenshots, has been published previously [].
All patients, in both the intervention and control groups, received standard care, which included an assigned contact nurse and a visit with the physician before treatment.
Data CollectionBefore randomization, patients self-reported sociodemographic characteristics, including education level, marital and occupational status, and baseline outcomes via questionnaires. In the B-RCT, follow-up (via postal questionnaires) occurred 2 weeks after the end of NACT or the day before surgery, whichever came first. In the P-RCT, follow-up was 3 weeks after the conclusion of RT. Medical history and clinical treatment data were obtained from the patients’ medical records, including comorbid conditions, tumor histopathology, cancer stage, and prostate-specific antigen score before treatment initiation, as well as the type and number of cancer treatments planned and completed, and reasons for discontinuing or altering treatment. Data on mortality and cause of death were obtained from the Stockholm and Gotland Regional Cancer Centre and the Swedish NBHW ().
Health Care Utilisation and CostsAdministrative data on each patient’s health care utilization and costs, from the first day of treatment and for 6 months thereafter, were obtained from the Stockholm Region Council administrative database (VAL). The database includes variables on primary care and emergency department visits (VAL-OVR) and hospitalizations (VAL-SLV) for Stockholm Region Council patients []. Health care costs were estimated using a variable (SIMKOST [simulerad kostnad/simulated cost]) that calculates the cost of visits based on the profit and loss account for the respective care branch. SIMKOST reflects approximately 90% of the costs for individuals’ visits to outpatient care and 99% of the costs for inpatient care (). Intervention costs were based on a fiscal estimate provided by the company that developed the app, expressed as a 1-time implementation/startup cost of €5212, with weekly licensing costs per capita of €39 for nurses and €2.25 for patients ().
Data AnalysisStatistical AnalysisData were handled using Microsoft Excel 2016 with the add-in XL-STAT, IBM SPSS Statistics version 27, and STATA 16 (StataCorp LP). Clinical trials with an RCT design should be analyzed using the intention-to-treat (ITT) principle [], so missing values were imputed as the mean per group and time []. Health care utilization and cost values were imputed for 2 patients in each intervention group (B-RCT and P-RCT). In the P-RCT, EORTC dimension scores were imputed at baseline for 2 patients per group, and at follow-up for 15 patients in the intervention group and 20 patients in the control group. In the B-RCT, follow-up values were imputed for 5 patients in the intervention group and 4 patients in the control group. Distribution normality was assessed using skewness and kurtosis. All costs were adjusted for inflation from 2019 to 2022 [] (×1.0764) and converted from Swedish kronor (SEK) to Euros (€) using the average exchange rate for April 2022 of 10.3257 SEK=€1 []. Nonparametric bootstrapping (1000 replications) was used to test nonnormally distributed variables, calculate the incremental cost-effectiveness ratios (ICERs), and explore sample uncertainty regarding the mean ICERs [].
Health OutcomeEORTC QLQ-C30 [] dimension scores were mapped onto EQ-5D-3L [] health state utilities using a response mapping algorithm [,] (). The original algorithm includes British utility weights [], which were replaced with Swedish weights according to Burström et al [] for this study. The mean predicted EQ-5D value (EQ-5DP) before treatment minus after treatment was used to measure effectiveness, with a smaller reduction in mean EQ-5DP indicating better outcomes.
Intervention CostsThe overall startup cost was divided by the total number of patients diagnosed and treated with the respective treatment regimens in the Stockholm Regional Council and Gotland Region for the years 2016-2018 (518 patients with breast cancer treated with NACT and 683 patients with prostate cancer treated with RT). Given that system updates may incur additional costs beyond the license fees, a time frame of 3 years was considered reasonable. The estimate assumed 5 nurses per 100 patients, with no additional costs for nurses to handle symptom alerts. Based on each patient’s number of weeks in treatment (wt), the intervention costs were calculated per equations (1) and (2) for B-RCT and P-RCT, respectively:
(5212/518) + ([39 × 5/100] × [wt]) + (2.25 × wt) (1)(5212/683) + ([39 × 5/100] × [wt]) + (2.25 × wt) (2)Cost-Utility AnalysisStochastic CUAs [] were conducted for each RCT by calculating ICERs in 3 different ways. For ICERa, each patient’s intervention cost, along with all health care costs from randomization through 6 months, was included. For ICERb, each patient’s intervention costs, plus acute health care costs from randomization and the subsequent 6 months, were considered. Given the considerable variation in patient health care costs, which introduced substantial uncertainty in the cost-effectiveness estimates, a third ICER (ICERc) was calculated by dividing the intervention group’s intervention costs minus the control group’s intervention costs by the difference in QALYs lost between the 2 groups. The rationale behind this approach is that patients’ health care utilization during cancer treatment is influenced by multiple factors, and a much larger study would be necessary to demonstrate a significant reduction in health care costs. Therefore, it was deemed appropriate to assess cost-effectiveness under the assumption that health care costs are not substantially affected. To capture the gradual change in the quality of life during treatment, QALYs lost were calculated linearly as follows: ([EQ-5DP after treatment minus EQ-5DP before treatment]/2) × (individual treatment duration in weeks/52). For visualization, bootstrap values of the incremental intervention costs and incremental health outcomes (QALYs) were plotted on cost-effectiveness planes.
In the P-RCT, the RT treatment was standardized with minimal variation between patients, so RT costs were excluded from both CUAs. By contrast, the B-RCT did not allow for standardized subtraction of treatment costs, so all health care costs were included. The analysis was conducted from the payer perspective (Stockholm Region Council) and focused on the patient’s treatment duration (less than 1 year), meaning that no discounting of costs or results was applied. The cost per QALY, as defined by the Swedish NBHW, was used to evaluate cost-effectiveness.
ICERa=[(intervention costs + IG total health care costs) – (CG total health care costs)]/(IG change in QALY – CG change in QALY)ICERb=[(intervention costs + IG acute health care costs) – (CG acute health care costs)]/(IG change in QALY – CG change in QALY)ICER=[(IG intervention costs) – (IG intervention costs)]/(IG change in QALY – CG change in QALY)Exploration of Health Care Utilization and CostsWithin each RCT, variables for total and acute health care visits and costs were generated by summing each participant’s visits and costs, conditional on the VAL variable AKUT (acute) being marked as yes or no. Additionally, variables for health care utilization related to the respective cancer treatments were created through a qualitative analysis of the International Statistical Classification of Diseases and Related Health Problems (ICD) codes, using a conventional and summative approach []. All ICD codes for acute outpatient and inpatient visits within each RCT were compiled in Excel sheets, and the occurrence of all unique codes was counted. These ICD codes were either grouped or coded based on similarities into predefined and emerging categories. Examples of these categories include fever/neutropenia (D709C, R502, R508, and R509), gastroenteritis/colitis (K521 and A047), anemia (D649), urinary tract infection (N390), and urinary problems (R339, N390, R301, N390X, N304, N109, T830, R391, R319, and N300; ).
Each patient’s visits and costs, according to the categories, were calculated to create variables used as dependent outcomes in multivariate regression analysis. Depending on the level of overdispersion, Poisson, negative binomial, or binary logistic models with a log-link function were fitted []. The variable “Group” was coded as control=0 and intervention=1. Prior studies have suggested an association between diminished performance status [,], the presence of multiple chronic diseases in older individuals [], and increased costs. Therefore, the continuous variables—age at inclusion, Charlson Comorbidity Score, and Baseline EQ-5DP score—were included as covariates. The reference category was arranged in ascending order. For the B-RCT, each patient’s number of NACT cycles was included as an independent variable. By contrast, for the prostate cancer trial, treatment was standardized, and all patients underwent a similar number of treatments.
The mean EQ-5DP before treatment was 0.86 in the intervention group and 0.87 in the control group. After treatment, the mean EQ-5DP was 0.84 in the intervention group and 0.80 in the control group (P=.036, effect size=0.099). A statistically significant difference was observed in the mean changes in EQ-5DP from before to after treatment between the intervention and control groups (P=.012, effect size=0.042). The greatest difference in change was observed in the Anxiety/Depression dimensions (). The CONSORT (Consolidated Standards of Reporting Trials) checklist is presented in (also see and ).
Cost-Utility AnalysisThe intervention group patients had a mean QALY loss of –0.004 (SD 0.002) from before treatment to after treatment, while the corresponding figure for patients in the control group was –0.012 (SD 0.002; P<.001). The mean cost for the Interaktor app per patient was €92 (SD €2). The mean total cost for the intervention and all health care was €36,882 (SD €1032) for patients in the intervention group and €35,427 (SD €959) for control group patients (P<.001). The ICERa was €202,368 (€SD 811,136; 95% CI €152,008-€252,728). The mean cost for the intervention and acute health care was €3585 (SD €480) in the intervention group and €3235 (SD €494) in the control group (P<.001). ICERb was €49,903 (SD €207,042; 95% CI €37,049-€62,758; ). Lastly, when health care costs were excluded from the analysis, the ICERc was €13,213 (SD €33,327; 95% CI €11,145-€15,281; ).
Table 1. Breast cancer trial cost-utility analysis.aIndependent unpaired samples Student t test (2-tailed).
bBased on bootstrap.
cQALY: quality-adjusted life year.
dN/A: not applicable.
e€1=US $1.03.
fICER: incremental cost-effectiveness ratio.
presents a cost-effectiveness plane depicting the bootstrapped values of the intervention group’s joint incremental costs and incremental QALYs compared with the control group, as per ICERa and ICERb. illustrates the cost-effectiveness plane with the corresponding values based on ICERc.
Figure 1. Breast cancer cost-effectiveness plane ICERa and ICERb. ICER: incremental cost-effectiveness ratio; QALY: quality-adjusted life year. 
Figure 2. Breast cancer cost-effectiveness plane ICERc. ICER: incremental cost-effectiveness ratio; QALY: quality-adjusted life year. Exploration of Health Care Utilization and Health Care CostsThe mean outpatient cost for patients in the intervention group was €27,571 (SD €6392), while in the control group, it was €26,348 (SD €5800). In both groups, approximately 2% of the outpatient costs were attributable to acute care: €554 out of €29,321 (1.9%) in the intervention group and €562 out of €26,348 (2.13%) in the control group.
In the intervention group, 13 out of 74 (18%) patients had an acute outpatient visit for fever, with a total of 34 visits. In the control group, the corresponding proportion was 9 out of 75 (12%), with a total of 21 visits. Additionally, 7 out of 74 (9%) patients in the intervention group had an unplanned admission from outpatient to inpatient care, accounting for 37 unplanned admissions. In the control group, 6 out of 75 (8%) had an unplanned admission from outpatient to inpatient care, totaling 29 unplanned admissions.
The mean inpatient cost per patient was €9207 (SD €5254) in the intervention group and €9093 (SD €5460) in the control group. Approximately one-third of all inpatient care cost was acute in both groups (€2932/€9207, 31.85% in the intervention group and €2665/€9093, 29.31% in the control group). The most common diagnoses during acute inpatient care episodes in both groups were fever, gastroenteritis/colitis, anemia, and urinary tract infection. The variable group (intervention/control) was not associated with the number of visits for fever, gastroenteritis/colitis, anemia, or urinary tract infection, nor were age, health-related quality of life (HRQOL) before treatment, comorbidities, or the number of NACT ().
Table 2. Breast cancer trial multivariate regression analysis of predictors for acute healthcare visits for chemotherapy-related symptomsaPearson χ2 value divided by degrees of freedom (goodness of fit of the model).
bNr of acute visits when a patient received the diagnose.
cN/A: not applicable.
dP value Omnibus test General Linear Model Negative Binomial Regression.
eIntervention/Control; Reference category=Intervention
fP value for the independent variable in the General Linear Model Negative Binomial Regression.
gNeoadjuvant chemotherapy treatments
Negative binomial multivariate regression analysis revealed that the independent variable group (intervention/control) did not significantly affect the predicted log odds of patients’ health care costs (P=.949). For acute outpatient health care costs, the analysis showed that age, health-related quality of life at baseline, and comorbidities significantly predicted costs. Older age (P=.002) and better health-related quality of life (P<.001) were associated with lower acute outpatient health care costs. By contrast, a higher number of comorbidities was associated with increased acute health care costs (P=.02; ).
Table 3. Breast cancer trial multivariate regression analysis of predictors for health care costs.
Comments (0)