This review will be reported in accordance with the standard PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines. The study has been registered with PROSPERO (CRD42024564759) [15] before the initial literature search [16]. The full planned search strategy is provided in the Supplementary Material in Multimedia Appendix 1, which has been developed in close concordance with a medical librarian (IK and EE). This search strategy includes the databases and data repositories: MedLine OVID, IEEE, SCOPUS, Physionet, National Sleep Research Resource, UC Irvine Machine Learning Repository, and Borealis.

We will include any peer-reviewed primary studies—observational and nonobservational—which use PPG datasets. This includes PPG datasets in published randomized controlled trials, nonrandomized trials, cross-sectional studies, cohort studies, and algorithmic models.

We will include studies published since 2000 to ensure the datasets are relevant to current technology developments. The year of 2000 was defined as the lower bound to ensure the datasets selected for analysis are aligned with contemporary medical device design, especially since there has been a rise in usage of PPG datasets to train wearable health technologies in the last 2 decades. In the early 2000s, significant global health milestones like the Millennium Development Goals were established, marking a period in time when critical frameworks and health care initiatives for promoting greater equity and diversity were implemented. We plan to include studies that mention PPG and specifically reference openly accessible datasets. Therefore, accounts will be created to access datasets as needed; however, any datasets that require a fee or monetary incentive to access the datasets will be deemed as not openly accessible. The datasets must contain physiological parameters such as heart rate, blood pressure, or respiratory rate as well as data describing the waveform, such as pulse wave or slope transit time. These terms have been chosen as they are indicators of parameters routinely collected in PPG datasets. The inclusion and exclusion criteria are outlined in Textbox 1. Included studies will be evaluated for the extent to which they include demographic characteristics and consider these characteristics.

Datasets meeting the eligibility criteria will be included for analysis of demographic characteristics. Articles mentioning openly accessible datasets may be used to provide contextual information on the datasets, but only the datasets will be analyzed to ensure the primary objective of this review is focused on the reporting quality of openly accessible PPG datasets.

Following the initial literature search, all studies will be reviewed independently by two authors using systematic review software, Rayyan.ai [17,18]. Studies will initially be screened by title and abstract. The full-text articles will then undergo screening. Disagreements will be discussed with and resolved by a third author, a senior author with expertise in the topic. At each step, studies will be retrieved only if they meet the eligibility criteria.

We predict that there will be heterogeneity in the data reporting quality and the terminology used to signify specific variables; therefore, it is important to use broad inclusion criteria to capture relevant articles.

We also anticipate significant heterogeneity of demographic characteristics. Jiang et al [19] reported that less than seven percent of physiological datasets, specifically those available on PhysioNet, include all 4 key demographic variables—age, sex, race, and ethnicity. Consequently, it is important to not include these variables in our inclusion or exclusion criteria to truly understand the level to which PPG datasets consider demographic aspects including variables such as age, sex, race, ethnicity, and other relevant parameters like skin tone or BMI. Understanding the extent to which demographic parameters are collected in PPG datasets will therefore shed light on what information is used in informing the development and deployment of healthcare technology. By collecting information on the structural aspects of the PPG datasets, we will provide a critical comprehension of the reporting quality of openly accessible datasets.

Understanding the reporting quality of datasets used to develop health care technology is becoming increasingly important as the use of algorithmic models and processing to train medical devices grows. By analyzing openly accessible PPG datasets, we aim to provide a more comprehensive view of standard of data collection and usage for medical devices. Because of PPG’s long-standing usage in medical devices since the 1970 s and the wide range of biosignals that can be extracted from PPG data, it is vital to increase transparency in the demographic information collected in tangent to the biosignal data. The principal finding from this review will be to extract the different demographic and biosignal characteristics listed in openly accessible PPG datasets. We will provide summary statistics including frequencies, proportions, and distributions of the number and types of dataset characteristics. These summary statistics will be visualized in diagrams like histograms and box-whisker plots as well as tables. Any additional information related to characteristics such as the geographical location of the data collection or socioeconomic status will be portrayed in heatmaps to provide a general overview of additional social variables.

Wearables including smartwatches are commonly used to track personal health measures such as heart rate, sleep, and blood pressure. These commercially available medical devices often use PPG to take physiological measurements. However, PPG data collected by wearables have been observed to be affected by skin tone and higher adipose tissue due to obesity [12]. In combination, differences of up to 61% in signal intensity have been observed [12]. The Fitzpatrick scale is commonly used to assess for differences in skin tone, and while it provides an insight into the physiological signal values caused by greater melanin levels, it is subjective and treats skin tone as a categorical, rather than a continuous, variable [22]. This has limitations when considering the many underlying factors that result in the chromatic appearance of an individual’s skin tone, particularly if complicated by clinical conditions such as jaundice or hypoxia. Exploring the extent to which variables such as this are captured within existing datasets may be instructive in understanding their inherent ability to account for overall skin coloration.

Clearly, this same need to account for diversity in skin tone and clinical condition applies to clinical PPG studies and their resultant datasets. In a clinical setting, having erroneous oxygen saturation values or heart rate measurements can result in poorer clinical outcomes. There is evidence that the obese and those with greater melanin content in their skin endure worse health outcomes, with inaccurate monitoring a clear potential contributor [22]. Obesity can be attributed to physiological changes such as the reduction in trans-epidermal water loss and an increase in skin thickness due to adipose tissue can influence the light scattering patterns needed to obtain a PPG reading [22]. In addition to melanin content and obesity, other factors influencing perfusion and skin temperature can all affect PPG. When expanding PPG technology to low-resource settings, there is a need to account for diverse populations not only in terms of melanation but also factors such as obesity and conditions altering perfusion. Moreover, they may have a greater diversity of endemic clinical conditions including anemia, parasitemia, and malnutrition—whose effects on PPG have not been clearly described. Effective clinical PPG monitoring in these patients mandates an interpretation based on data from representative populations.

This understanding of demographic and biosignal variables is critical to presenting the current standard of reporting quality across openly accessible PPG datasets. In a recent review of PhysioNet datasets, a data repository with biosignal datasets, information regarding demographic and geographical information was presented to highlight the disparities and demographic gaps present in biosignal datasets [19]. This marks the analysis of openly accessible PPG datasets relevant and important for promoting transparency and equity in medical device development, as PPG is a widely accepted technology growing in popularity in health care. The aims of this systematic review mirror those of the review by Jiang et al [19], but take a broader look at openly accessible PPG datasets across multiple databases and repositories.

Providing a detailed summary of the demographic and biosignal variables in openly available datasets, we aim to promote a more comprehensive understanding of reporting standards in existing PPG datasets. This review will provide a structured overview of the current gaps and sources of bias in developing PPG-based medical devices. Transparency in dataset reporting will highlight the need for greater diversity and promote an increased range of demographic characteristics collected in openly available PPG datasets. However, we do expect limitations such as heterogeneity in the datasets, especially as it comes to which characteristics are collected and included in the datasets. We also anticipate challenges in gaining access to certain datasets as we will likely have to go through repositories or databases that require account creation or multiple steps in gaining access to the PPG databases. These limitations will be mitigated by maintaining clear documentation of the characteristics extracted from the datasets being analyzed and data harmonization. Also, accounts will be created as needed to ensure adequate access is gained to review the dataset characteristics.

In this systematic review, we will analyze the structural basis of PPG datasets and critically evaluate the different variables included in the datasets. This analysis will highlight the reporting quality of openly accessible PPG datasets. By critically appraising the dataset diversity and reporting quality, we can highlight the gaps present in current PPG data to inform future collection and design of medical devices including wearables to be more inclusive to avoid perpetuating biases. In addition, by looking at the structure of PPG datasets and their representativeness, we can also provide insight into the gaps that may be present in the context of PPG applications in low-resource settings. The findings can promote awareness of sources of bias in openly accessible PPG datasets, informing future data collection and modulation when training PPG-based medical technology. With the synthesis of this information and data analysis, we hope that medical device developers would consider dataset level biases and their implications on human health. We hope that these considerations would push toward greater transparency in PPG data collection and extraction of demographic and biosignal characteristics to ensure users are informed of the design and development of medical devices and avenues for promoting fairness. This review, therefore, aims to address disparities in openly available PPG datasets, the downstream effects of these disparities, and potential avenues for greater equity in PPG dataset creation and usage.