To capture QoL trajectories (QoL-T) with high resolution, that is, biweekly, weekly, or even daily, each patient used their own mobile device for data collection. The alternative would be a preinstalled device provided by the health care organization, which is associated with significant cost and logistical effort and therefore cannot provide a scalable system. Patient devices were very heterogeneous: smartphone or tablet with iOS or Android, laptop or PC, and different operating systems with different software versions. Therefore, a web-based app (HTML5) was developed for data collection, called MyEDC (my electronic data capture), which is based on OpenEDC [9,10]. This approach avoids software installations by the patient.

Figure 1 presents the high-level IT architecture for data transfer from the patient device to the EHR. This architecture was reviewed and approved by the information security officer of Heidelberg University Hospital (UKHD). The web app was provided by an electronic data capture (EDC) proxy in the demilitarized zone, which is protected by 3 firewalls and a reverse proxy. In the internal network, a job service (programmed in Java with Spring framework) sent emails with 1-time EDC links to the patient. The job service polled data from outside to the internal network but only for registered and valid data collections. Other data packages were ignored and deleted. Valid data were transferred to an internal EDC database, which converted data into EHR format and performed an automated upload to the EHR database. A Clinical Data Interchange Standards Consortium (CDISC) operational data model [11]–compliant EDC system was used to comply with data standards in clinical trials. Owing to technical limitations of the local EHR system (i.s.h.med from Cerner Inc), files in PDF with informative document titles were generated and uploaded into the EHR.

Overall, HRQoL was assessed frequently using a visual analog scale (VAS) with a range from 1 to 100 (Figure 2), similar to HRQoL instruments such as EQ-5D. An HRQoL value of 0 was reserved for deceased patients. QoL-T, that is, a time series of HRQoL values, were plotted and summarized with a normalized area under the curve (AUC) per patient. Few (mostly <5) additional, disease-specific PROM items were added for each clinical use case, for example, pain in surgical patients. Simple language was used to support a wide variety of use cases, including severely ill patients and participants with limited language skills. These questionnaires were designed by clinical experts to assess patient trajectories regarding effectiveness and side effects of routine therapy.

Statistical analyses were performed using R (version 4.4.0). Differences between paired observations were assessed using the Wilcoxon signed-rank test. Associations between continuous or ordinal variables were investigated using the Spearman rank correlation coefficient. For all tests, a 2-sided P value threshold of .05 was considered statistically significant. AUC was calculated with the trapezoidal rule using the mean of adjacent points, normalized by total time, and scaled to a value range from 0 to 1. A 95% confidence band for average QoL was calculated as a point-wise 95% CI of the mean. The participation rate was defined as the percentage of eligible patients who consented to participate in the study out of the total number of individuals invited. To assess longitudinal data quality, we calculated the RR at each assessment time point, defined as the percentage of participating patients who completed the questionnaire at that specific interval. Time to answer was calculated for completed questionnaires. Clinical feasibility for QoL-T was assessed regarding completeness with the median RR across all time points. Timeliness of patient answers to short PROM questionnaires was described with median time to answer.

This study was conducted in accordance with the ethical guidelines of the Declaration of Helsinki. The research was approved by the ethics committee of the Medical Faculty of Heidelberg University (S-607/2023 and S-255/2025). The privacy and confidentiality of participants were maintained by ensuring that all study data were deidentified. Informed consent was obtained from all participants; this included the explicit right to opt out or withdraw from the study at any time. No direct compensation was provided for participation.

At present, survival (overall or progression free) is the most established primary end point in clinical research, especially in oncology [19]. Survival has the advantage that it can be collected with high precision in time (day of death) and completeness. However, survival is a crude measure for therapeutic success because it ignores PROMs such as QoL. Therefore, PROMs are increasingly used as end points in clinical studies [20]. This approach also has limitations because it only assesses the outcome at the end of an observation interval but not the patient trajectory: when 2 therapies result in the same average HRQoL after 1 year, therapy A can still be superior to therapy B, if HRQoL is significantly better during the time course of this year. Many established HRQoL instruments such as the WHO-5 well-being index, Functional Assessment of Cancer Therapy-General (FACT-G), or EQ-5D have short recall periods (<2 weeks). Consequently, frequent data collection (every 2 weeks up to daily) is needed for complete, high-resolution HRQoL trajectories. This is a methodological challenge because low RRs were reported in several settings.

Our hypothesis was that assessing HRQoL on patient’ mobile devices and integrating these findings into the EHR system can enable such high-resolution collection of HRQoL because a patient would be motivated to contribute to their own therapy. In our setting, a patient could directly inform their treatment team about QoL and symptoms. We could show that this approach is actually working in 4 completely different medical domains (psychosomatics, hematology, visceral surgery, and neurosurgery), both from a technical and a clinical perspective. Clinical staff emphasized the value of EHR integration because it supports patient care, for example, regarding pain management. Our proposed IT architecture proved itself in pilot operation. Patients were responding to these HRQoL requests on a regular basis.

What to measure regarding patient outcomes, including the frequency of assessment, was described as a key challenge [8]. We applied a 100-level VAS with a question in simple language because, from a clinical point of view, long texts are not suitable for seriously ill patients (eg, intensive care setting). It was demonstrated that even for those patients (eg, directly after major surgery), a large proportion is able and willing to answer HRQoL questions with their smartphones on a daily basis. In the European Union, software that calculates a clinical score may be classified as a medical device, depending on its intended use and clinical impact. While a single-item HRQoL question may be viewed as “clinical documentation,” any software-based assessment intended to influence diagnostic or therapeutic decisions requires rigorous validation. Given the vast technical heterogeneity of patient-owned smartphones, the technical and clinical validation required for such software to meet medical device standards remains a significant hurdle for large-scale implementation. More complex HRQoL instruments, such as the EORTC QLQ-C30, can provide a more detailed assessment of different QoL dimensions but require substantially more documentation effort. We propose to use such instruments in larger time intervals combined with high-frequency short HRQoL assessments. In our study, patients of almost all age groups were capable to do this data collection. The clinical team or family members can support patients who are not able to handle a mobile device.

The median time from email to answer was <8 hours; therefore, daily or weekly assessments are feasible. We propose starting the VAS with 1 and assigning an HRQoL value of 0 only to patients who are deceased. By this means, HRQoL values can be directly transformed into survival data (dead=HRQoL 0 and alive=HRQoL 1-100). This enables a direct comparison of HRQoL trajectories with survival, which is the current standard end point.

PROMs integrated with EHR were described previously [21]. High-resolution HRQoL trajectories within the EHR constitute a novel readout for the course of disease. This level of detail was previously unavailable and, to our knowledge, has not been comprehensively established for any disease. We demonstrated that this approach is feasible across multiple, clinically distinct domains, ranging from psychosomatics to visceral surgery. HRQoL trajectories could support early detection of complications. In contrast to survival analysis, effects of interventions such as chemotherapy or surgery on QoL can be assessed quantitatively with HRQoL trajectories. These trajectories can potentially enable sensitive comparison of different medical interventions in the future and change the course of therapy.

A related method is ecological momentary assessment [22], which captures real-time data about patients’ experiences and behaviors in their natural environment. A similar approach such as in our pilot study was reported with the eRAPID system [23], which was applied for patients with breast, gynecological, and colorectal cancer. With eRAPID, the patient logged onto a web-based system (Qtool) to complete online questionnaires. To simplify this step, we sent a 1-time EDC link to the patient (ie, patient does not need an individual password). We used a CDISC-compliant EDC database to make the approach more scalable: the Medical Data Models portal [24] provides ≥25.000 questionnaires for download, which can be used for data collection with CDISC-compliant systems. Of note, the eRAPID team from Leeds was able to demonstrate that PROM collection as intervention can improve patient outcomes in patients with tumor (physical well-being and self-efficacy) [25]. Improved patient outcomes induced by PROM collection was also reported in other medical domains such as orthopedics [26]. In the field of cardiology, it was reported that a disease-specific PROM score is the strongest predictor for survival of heart failure [27].

Our findings are based on a single site pilot study with a relatively small sample size for each clinical use case. Consequently, our analysis may have lacked sufficient statistical power to detect smaller effect sizes. Due to technical constraints of the local EHR system, integration was limited to PDF files. Fast Healthcare Interoperability Resources–based EHR integration can enable improved support for clinical workflows. Further research is needed to define the best time intervals and most suitable assessment instruments for HRQoL collection. Most probably, these parameters are disease-specific. In some use cases, HRQoL measurements can be extended to the period before therapy. The rate of patients who accepted the electronic system varied considerably between use cases: in psychosomatics, >90% agreed to participate. In contrast, in visceral surgery, only about one-third of patients were willing or able to collect HRQoL data on the smartphone in the perioperative phase. HRQoL has several dimensions (eg, physical and psychosocial QoL). To compare 2 treatment schemes regarding HRQoL, a reduction to 1 dimension is needed (is therapy A better than B?). Therefore, further research is needed to define the most appropriate calculation of overall HRQoL. Another research topic is the metric to aggregate a series of HRQoL values. In this pilot study, we proposed AUC of HRQoL over time. Patient questionnaires can have a significant nonresponse bias, for example, language proficiency has a significant effect on RR [28]. In our setting, some patients had no access to a smartphone. Therefore, additional data capture methods (eg, tablet and interview by a health care professional) are needed for complete HRQoL data. At present, smartphones are used by >95% of the population in high-income countries [29]. Electronic HRQoL collection software needs to be further improved, given the disease-related limitations for patients compared with the general population. Software with high usability can contribute to address this problem: for instance, the study by Carter et al [30] reported an increase in PROM completeness from 24.9% to 67.0% by user-centered software design. Future work will focus on leveraging HRQoL trajectories to provide clinical decision support and enabling physicians to personalize therapeutic interventions, while simultaneously establishing a sensitive, dynamic outcome measure for future clinical trials.

High-resolution QoL-T with EHR integration are clinically feasible in a broad range of clinical settings. Collection of HRQoL data via patients’ mobile devices and secure data transfer to the protected EHR is technically feasible. These trajectories can provide a novel readout, which is not covered by survival analysis or a PROM end point. These data can potentially be useful to characterize diseases and compare different therapies with high sensitivity and time resolution.