We conducted this systematic review in accordance with a predefined protocol. The review was reported in line with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) statement (Multimedia Appendix 1) [15,16]. The review was registered at PROSPERO (International Prospective Register of Systematic Reviews; registration number CRD42021226896; date December 18, 2020).

Only studies written in English were included. Studies selected for inclusion were found to meet the following PICOS (Population, Intervention, Comparison, Outcomes, and Study Design) framework (Table 1) [15]:

The research team developed the search strategy. We sought advice from an information expert at the Royal Library to ensure all relevant Medical Subject Headings (MeSH) and filters were addressed. We conducted several pilot searches to capture targeted papers. The underlying principle of the search strategy is delineated in Textbox 1. The specific strategy and search strings are presented in Tables S1-S4 in Multimedia Appendix 2. We systematically searched the following electronic databases for potentially eligible papers: PubMed/MEDLINE, CINAHL, EMBASE, and Web of Science. The initial search was completed on December 18, 2020, for all databases and repeated on February 8, 2023, for PubMed/MEDLINE only (Table S5 in Multimedia Appendix 2).

Titles and abstracts of the identified papers retrieved from electronic databases and other searches were exported to the systematic review management software Covidence, which was used solely to manage references and remove duplicates. No automated screening was performed in Covidence. Two reviewers independently screened all titles and abstracts according to the eligibility criteria. Full texts were reviewed if eligibility could not be determined by the title and abstract alone or in the case of disagreement. We made the final decision in agreement by going through all the included full texts in the group. The study selection process was reported in a flowchart.

Risk of bias was assessed by one of three of the authors independently using version 2 of the Cochrane risk-of-bias tool for randomized trials (RoB 2) [18]. The assessment was hereafter reviewed by one of the other authors. In the case of discrepancies, the risk-of-bias assessment was agreed on in a joint evaluation. One author also authored one of the included studies and was not involved in the risk-of-bias assessment.

Data synthesis followed the synthesis without meta-analysis (SWiM) guidelines [19]. In short, the key characteristics of the studies were extracted and recorded in a predesigned table. If an intervention was reported in 2 or more papers, information about the first author and publication year from all papers were collapsed and reported under the primary publication. We used information from all papers to report the intervention and primary and secondary outcomes. Based on the extracted information, we created a narrative synthesis presenting the characteristics of the studies, the PROM characteristics, and the outcomes. The studies were grouped according to primary outcomes and study populations, and since the outcomes varied in content and type, data were synthesized with vote counting based on direction of effects. To ascertain the characteristics of the most effective PROM interventions, we categorized studies with a low or moderate risk of bias, reporting a positive effect if the intervention group demonstrated a statistically significant superior outcome compared to the control group or reporting noninferiority if the difference between the intervention and control groups was below the noninferiority threshold. The characteristics of these studies were synthesized separately. The PROSPERO protocol included assessment with GRADE (Grading of Recommendations Assessment, Development and Evaluation) [20]; however, this method was discarded because of intervention and outcome heterogeneity, so consistency of effects could not be assessed. More study details are described in Tables S6 and S7 in Multimedia Appendix 2.

The search identified 31,902 records. After identification of duplicates, 21,203 (66.5%) papers that were screened at the title/abstract level resulted in 686 (3.2%) potentially eligible papers. Following full-text assessment, 606 (88.3%) reports were excluded, mainly due to wrong interventions and publication types, and 56 trials described in 79 (13%) papers [21-99] were included (see the flowchart in Figure 1). The number of study participants varied widely (range: 17-1653 individuals, median 212.5 individuals). Of the 56 trials, 17 (30.4%) included patients with heart disease, 13 (23.2%) included patients with COPD, 10 (17.9%) included patients with diabetes, 9 (16.1%) included patients with IBD, 6 (10.7%) included patients with RA, and, finally, 1 (1.7%) included patients with mixed diagnoses (heart disease and diabetes).

Characteristics of the selected studies and PROMs are summarized in Table 2. Telephonic PROM administration was most common, followed by devices and applications (apps). The weighted mean age among participants receiving a PROM intervention delivered over the phone was 62.9 years (63 years for phone calls and 57.9 years for SMS). The first described use of devices for collection of PROMs was published in 2008. Apps for tablets and smartphones were introduced in 2007 and 2009, respectively. The participants’ weighted mean age was higher in interventions that used devices (71 years) than interventions using apps, webpages, or laptops (53.2 years).

All interventions included systematic PROM assessments, and some additionally included a predefined method for PROM-based decision-making (n=19, 33.9%) [21-39] or PROM-based dialogue (n=9, 16.1%) [40-48], and 5 (8.9%) interventions intended to replace face-to-face visits [49-53]. The PROM interventions used validated scales in 23 (41.1%) of the studies, while specified questions or unspecified questions were used in the remaining interventions. The content of the PROMs varied from symptoms and health status to the HRQoL, health beliefs, and self-care. All interventions included feedback to HCPs as per the inclusion criteria, while 23 (41.1%) also provided feedback to patients [22,27,28,30,32,33,36,37,39, 41-43,50,51,54-62]. In 39 (69.6%) of the interventions, predefined cutoff values were used to ensure feedback to the HCPs, 18 (46.2%) of which included alerts. Intervals between PROMs were not always documented but varied widely from a single assessment to collections twice daily. In the 20 (35.7%) interventions where PROMs were collected daily or twice daily, the number of questions varied from 1 to 15. Daily collection was used for COPD (n=9, 16.1%) [28,32,34,35,43,55,63-65], heart disease (n=9, 16.1%) [36,37,57-59,61,66,67], IBD (n=1, 1.8%) [23], and RA (n=1, 1.8%) [68] treatment. The daily PROM collection continued for 3-24 months. Studies only collecting PROMs once or with long intervals between had more comprehensive questionnaires.

Characteristics of the studies and interventions are presented in Tables S6 and S7 in Multimedia Appendix 2.

All trials included more than one outcome. The following review of the results focuses on primary outcomes (Table S8 in Multimedia Appendix 2). Secondary outcomes are listed in Table S9 in Multimedia Appendix 2. Of the 56 trials, 12 (21.4%) had favorable results in the primary outcomes, 3 (5.4%) were noninferior, 6 (10.7%) had more than one primary outcome and mixed results, 6 (10.7%) were feasible, 1 (1.8%) was not feasible, 2 (3.6%) had no group comparison, 24 (42.9%) had no effect, and 2 (3.6%) had negative results by increasing hospital admissions with the intent of reducing them.

Of the 10 (17.9%) studies of patients with diabetes, 1 (10%) concluded noninferiority [53], 1 (10%) focused on feasibility [40], 1 (10%) demonstrated a mixed effect of PROMs [44], and the other 7 (70%) studies showed no effect [30,31,33,45, 47,69,99]. For studies of patients with COPD (n=13, 23.2%), there was a positive or partially positive effect in 6 (46.2%); in addition, 2 (66.7%) of 3 feasibility studies were feasible [64,65]. In the 18 (32.1%) studies including patients with heart disease, 5 (27.8%) were positive and 1 (5.6%) intervention was feasible [59]. There was a positive effect of the intervention in 33% of the studies for patients with IBD (n=9, 16.1%), 2 (22.2%) were feasible [23,39], and 1 (11.1%) was noninferior [52]. Half of the studies including patients with RA were positive, and 1 (16.7%) intervention was noninferior [50].

Several studies indicated adverse events [23,38,39,49,61, 65,71,72,89]. These included 2 (22.2%) studies in patients with heart disease and diabetes, which intended to reduce hospital admissions but increased the number instead [61, 72], and 1 (11.1%) study in which the subgroup analyses indicated increased mortality in women [89]. Adverse events were also observed in secondary outcomes: decreased peak performance and physical activity [65], increased hospital admissions [38], increased medical treatment [39], deterioration in symptoms [71], increased symptoms of depression [23], and higher costs [49].

The predefined secondary aim of this systematic review was to describe effective PROM interventions with acceptable quality defined by a low risk of bias or some concerns. Excluding a study [24] that was later debunked by a larger study [89], 6 (10.7%) studies had positive results in the primary endpoint, had positive results in one of two primary endpoints, or were noninferior in replacing face-to-face sessions [25,42,46,50,53,57]. The PROM interventions used multiple administration forms and were focused on symptoms of disease, and 5 (83.3%) of the 6 studies included support for decisions and dialogue [25,42,46,50,53], 4 (66.7%) studies included validated PROMs [42,46,50,53], and 2 (33.3%) included specified questions [25,57]. In addition, 5 (83.3%) of the 6 had 2-week to 6-month intervals between PROM collection [25,42,46,50,53], and 4 (66.7%) of the 6 studies included feedback to patients as well as HCPs, with 3 (75%) having a graphical display of the results [42,46,50]. Furthermore, 2 (33.3%) studies included patients with heart disease [25,57], 2 (33.3%) included patients with RA [42,50], 1 (16.7%) included patients with diabetes [53], and 1 (16.7%) included patients with COPD [46]. Positive effects were achieved in the HRQoL [46], physician-reported interaction [42], self-efficacy [57], a composite endpoint with hospital admissions and death [25], and noninferior RA or diabetes activity when replacing outpatient visits with PROMs [50,53]. These 6 (10.7%) studies had little loss to follow-up, and 5 (83.3%) of the 6 studies had more study consenters than study decliners [25,42,46,50,53].

Seven studies were rated with an overall low risk of bias (Figure 2). Among the 22 (39.3%) studies with some concerns regarding risk of bias, concerns relating to the reporting of results were the most frequent (n=13, 59.1%), followed by concerns arising from deviations from the intended intervention (n=10, 45.5%), the randomization process (n=9, 40.9%), the measurement of the outcome (n=5, 22.7%), or missing outcome data (n=5, 22.7%). Nearly half of the studies had a high risk of bias (n=27, 48.2%). No studies had biases in all domains, but 2 (3.6%) studies had high risk in four of five domains [27,38]. The most frequent domain for a high risk of bias was selection of reported results (n=18, 32.1%) led by a lack of prospective descriptions of the trials in registries or protocol papers. Among studies with a low risk of bias or some concerns, 14.3% (4/28) had positive results, while among studies with a high risk of bias, 33.3% (9/27) had positive results. Risk-of-bias evaluation of the individual studies can be found in Table S8 in Multimedia Appendix 2.

This systematic literature review identified sparse evidence regarding the effectiveness of the clinical use of PROMs in patients with five noncommunicable diseases: diabetes, COPD, heart disease, RA, and IBD. Considering the increase in PROMs’ popularity and the fact that we included five highly prevalent and burdensome diseases [100-102], it is notable that the literature search only revealed 56 trials in total, half of which had a high risk of bias. The included RCTs investigated a wide range of outcomes, thus producing a heterogeneous set of data and indicating a lack of consensus on the role of PROMs in clinical practice. Of the 56 RCTs, 12 presented positive results in the primary endpoint, 6 had mixed results in multiple primary endpoints, and 3 were noninferior. Focusing on the 28 RCTs with a low risk of bias or some concerns, 4 resulted in a better primary outcome and 2 demonstrated noninferiority. Successful interventions were characterized by using validated or specific PROMs and had little loss to follow up. They provided feedback to patients as well as HCPs, with 3 interventions incorporating a graphical display of the results.

Compared to the results of this review, the results of reviews about acute cancer treatment are summarized as positive [14,103]. However, the majority of the included studies lacked power to detect group differences, and the conclusion of predominantly positive findings may be questioned [14]. The conditions of acute cancer treatment may also differ from those of long-term diseases where patients may have had more time to recognize, to learn self-management of symptoms and side effects, and to seek relevant health care.

This review was conducted with screening by two independent researchers, and discrepancies were handled by including a third researcher in a consensus approach to ensure internal validity. The limited descriptions of interventions posed certain constraints, often lacking detail, which may have affected the external validity of this review. Adopting a more conservative approach of including only pure PROM interventions with no objective measurements at all would have reduced the number of eligible papers and the clinical relevance. Conversely, a less stringent approach, including more interventions where PROMs were one of multiple components, would have reduced focus and confidence in the results. Second, the heterogeneity of the studied PROM interventions and outcomes prevented meta-analyses and permitted only vote counting for data synthesis, which does not allow for differential weights to be allocated to each study.

The proportion of papers included in the review to the original number of studies identified in the search was less than 0.3%. The likely cause was the intervention search words that did not allow for distinction between RCTs using PROMs as endpoints and RCTs using PROMs for intervention. Attempts to reduce the number of papers for screening would have excluded relevant papers, which is why the searches were placed at a high level of sensitivity. In comparison, the systematic review by Graupner et al [14] about PROMs focusing on cancer revealed a quite similar proportion of included studies as 22 of 8341 identified studies were included.

The expectations toward PROMs are underlined by the implementation activities that have preceded robust evidence. Implementation before evaluation demonstrates the dissimilar approach toward PROM interventions to that of drug interventions that would have required robust trials before being released into the market. As demonstrated in this review, adverse effects may also occur in PROM interventions. Furthermore, PROMs require time and resources of HCPs as well as patients, and robust evaluations of PROMs are urgently needed.

PROMs are expected to support a person-centered approach. With 3 noninferiority trials, most trials evaluated superiority in a wide range of endpoints. The range of endpoints underscore the unclear expectations toward PROMs and the difficulty in operationalizing patient-centeredness. Noninferiority trials may be preferable if the purpose of PROMs is to replace face-to-face outpatient visits to compensate for the resources spent on PROM administration and implementation. However, trials to demonstrate noninferiority with a clinically acceptable margin may not be feasible to conduct [104]. Furthermore, most PROM interventions are designed as an addition to usual care, in which case only demonstrated superiority would merit spending resources on PROMs.

The 6 effective PROM interventions may serve as inspiration for further development. They focused on symptoms and were mostly used for systematic PROM assessment, plus a predefined PROM-based decision support method or dialogue support as opposed to systematic PROM assessment alone. This underscores the significance of the organizational structure to ensure consistent monitoring of PROMs and suitable predefined clinical actions. The effective PROM interventions had little loss to follow up and few study decliners, which is a testament to the acceptability of the interventions. A review of patients’ reasons for not using digital PROMs has demonstrated the significance of patients’ health (patients would forego PROMs if they were too well or too sick) and that it could be burdensome to be confronted with one’s poor health and that technical problems, a lack of skills, language problems, and uncertainty about data security could all act as barriers toward the use of digital PROMs [105]. Patients with a moderate symptom burden who use PROMs could be referred to health care services at the expense of the most burdened patients who opt out of using PROMs, thereby increasing health care inequality.

Adding PROMS to usual care does not appear superior; however, replacing specific face-to-face visits with PROMS seems noninferior in patients with diabetes [53], IBD [52], and RA [50]. Noninferiority trials may be more suitable when PROMs are intended to replace face-to-face outpatient visits, thereby offsetting the resources allocated to the administration and implementation of PROMs.

The evidence supporting the clinical use of PROMs in the five chronic diseases included in this study is limited. In these diseases, the clinical application of PROMs does not seem to offer any advantage over standard care in terms of any endpoint. To derive insights from successful research, we highlighted 6 PROM interventions that demonstrated positive effects and exhibited a low risk of bias. These were characterized by symptom assessment with predefined cutoffs used for decision and dialogue support.