Secondary use of health data is defined by the American Medical Informatics Association as “non-direct care use of PHI [personal health information] including but not limited to analysis, research, quality/safety measurement, public health, payment, provider certification or accreditation, and marketing and other business including strictly commercial activities” [24]. On the basis of a systematic scoping review, Robertson and colleagues categorized secondary use by distinguishing between four types of secondary use of clinical data: research, improving quality and safety of care, informing financial management, and education [25]. These and other similar classifications [26-28] help in understanding the broad spectrum of secondary uses of clinical data and illustrate that it is not limited to research. However, as pertains to the focus of this paper, that is, the secondary use of clinical data in data-gathering, non-interventional research or learning activities, the generic classifications lack further specifications and detail.

The term secondary use (or reuse) implies that there is also a primary use. Primary use encompasses the generation and use of data within the context of individual health care in hospitals and doctors’ offices to serve direct care needs. Secondary use refers to the use of these data for purposes other than individual care. Two points of criticism could be made against the distinction between primary and secondary use and the implied distinction between care and research. First, the distinction between care and research is notoriously difficult and widely discussed. Second, if secondary use of clinical data was to be implemented in the future as a standard in the health care system (ie, all health care data would be made available for possible research), this could lead to a tendency to collect data more systematically or collect more data than necessary in the care context, thereby blurring the distinction between data generation for care (primary use) and data generation for research (secondary use). In arguing against these two criticisms, however, we find that (1) ethical and legal codes are still fundamentally based on the distinction between care and research, which is unlikely to be abandoned in the coming years as blurred boundaries in governance and regulation are difficult to manage. Furthermore, we argue (2) that even if data are collected with the additional motivation of secondary use, the generation and collection of data would be driven by the primary goal of individual care.

Clinical data refers to data generated and collected in clinical contexts for patient care (diagnoses, anamnesis, treatment, medication, and so on) as well as data for accounting purposes or patient management, such as age, employment status, and other sociodemographic information. Clinical data also include data generated in the course of nonroutine treatment (off-label or experimental therapies) as long as it is collected for the primary objective of individual care. In terms of clinical data, we also include data reported from health care services and units to health insurance. In contrast to the widely used term health data, clinical data includes neither data gathered by (common, ie, nonclinical) smart devices and smartphone apps or research data understood as data generated for research purposes.

In contrast to research that generates data, SeConts only gathers existing clinical data. The gathering of data in SeConts can be done either by collecting clinical data of different patients from one institution or several institutions or by collecting only the results of analyses of clinical data carried out in the institution of primary use (decentralized analysis).

SeConts is non-interventional as it does not use data from interventions carried out with the aim or priority of data generation for research. SeConts solely uses existing data from medical care. The data used in SeConts may originate from interventions, but only from interventions carried out for the sake of individual diagnosis and treatment.

Both the terms research and learning activities refer to investigations in which the acquisition of generalizable knowledge beyond the needs and logic of individual care is the sole or primary intension. Research and learning activities cannot be clearly separated, but a rough distinction can be made in terms of their respective objectives. Research primarily aims at acquiring scientifically generalizable knowledge to be shared within the scientific community through scientific publications. Research usually indirectly improves health care and is realized through publications and by implementing practical conclusions based on research results. Learning activities, on the other hand, are designed to acquire knowledge about current care practices (eg, in a defined care unit) to derive appropriate and immediate measures to directly improve a specific health care unit or service. When talking about improving health care, the question arises whether SeConts has the potential to directly benefit the individual patient whose clinical data are reused. A direct (therapeutic) benefit is only possible under certain circumstances, for example, infection control of a clinical unit or research on a chronic disease. Younger patients with such a chronic disease could possibly benefit in several years from research that improves treatment of that very disease. However, as such benefits seem rather unlikely and exceptional, we understand SeConts as activities that are neither intended nor expected to directly benefit the individual patient whose data are used. In addition, there are ethical reasons for this narrow definition of SeConts, namely, to avoid any possibility of therapeutic misunderstanding.

In summary, the central concept of this paper, that is, SeConts can be defined as activities that:

In a nutshell, SeConts describes activities that solely use data produced for the sake of health care and in the context of health care to improve biomedical science or services.

Having clarified the concept of SeConts in the previous section, we now examine the scope and practical relevance of the concept of SeConts as previously defined. Which studies fall under the concept of SeConts (scope)? What is their practical relevance to medical research and improving medical care? We explore these questions in the next section. The literature review carried out to investigate the scope and practical relevance of research or learning activities that fall within our definition of SeConts (step 1) led to the following results: many types of research or learning activities common in the field of quality control and quality improvement can be subsumed under the concept of SeConts (in particular, under the term learning activities). Examples include improvement of infection control, which can be done by monitoring clinical data in hospitals to identify patients at high risk of infection [29]. Clinical data can also be used to create computerized algorithms for the early detection of possible hazards from germs [30]. These activities can be considered a component of the comprehensive ideal for transforming a particular health care institution into an LHCS.

At the national level, clinical data are used for public health surveillance. For instance, data from EHRs are searched for indicators of influenza in primary care to detect a pandemic in its early phase [31]. Epidemiological studies focus on the distribution of diseases as well as their causes and effects in populations, such as studies on the epidemiology of a certain infection to inform and improve vaccination initiatives [32]. In outcomes research, the effects (outcome) of certain interventions are investigated, such as the effects of a nationwide antismoking law on childbirth in the area of public health [33] or, on a clinical care level, evaluating the quality of care [34]. In health services research, investigators can use clinical data to explore the mean costs associated with (treatment of) a certain disease [35]. A well-established form of secondary use of clinical data is registry studies analyzing collections of data on all patients affected by a particular disease (registries) such as cancer registries [36].

Clinical data are also reused in clinical research in the form of in-silico hypothesis testing, where clinical trials are modeled with the help of data from EHRs [37]. Moreover, clinical data can be used for comparative effectiveness research [38] to compare “the benefits and harms of alternative methods to prevent, diagnose, treat, and monitor a clinical condition or to improve the delivery of care” [39]. Evaluation of experimental therapies can be conducted in terms of SeConts, for instance, by sharing data from single off-label (or compassionate) use from different hospitals. The secondary use of data from these therapies can help inform other physicians with similar patients about the course and outcome of different experimental therapeutic approaches [40]. Other studies that reuse clinical data in the sense of SeConts are drug safety and efficacy studies [41]. In addition, some basic research is conceivable using existing clinical data, such as studies on risk factors for certain diseases that can be linked to single influencing or moderating factors [42,43]. Another form of secondary use of clinical data lies in the area of informatics research, which uses clinical data to develop new software tools that have the potential to improve patient care or analyze and improve data security within a health system [44].

In addition to the different studies mentioned above, clinical data can be used in an explorative manner, which can be understood as encompassing three (potentially subsequent) steps. First, data can be analyzed (possibly using artificial intelligence) to generate hypotheses for future studies; second, in the case of new research ideas resulting from this, the data can be used to check feasibility; and third, to identify potential participants for recruitment of upcoming studies [45].

To further clarify the scope and concept of SeConts (by means of definitio ex negativo), we also want to mention two areas beyond the concept of SeConts: prospective clinical trials (eg, randomized controlled trials) and all other forms of studies that include interventions carried out for the sake of generating data; genome wide association studies in so far as they combine clinical data (phenotypes) with sequencing data generated for research purposes (genotypes).

Thus far, we have illustrated that there are many types of research or learning activities that can be conducted in the sense of SeConts. These types of research or learning activities display huge heterogeneity with respect to study designs, research questions, and contexts. Even the names of the types show this heterogeneity, as they refer to very different attributes describing the respective type of SeConts: some names refer to a specific method or study design (eg, in-silico hypothesis testing), whereas others refer to a subject area (eg, epidemiology). Given this heterogeneity, the question arises as to whether the different types of SeConts can be categorized in a way that allows for a better overview. The literature that attempts to capture the potential applications of secondary use of clinical data does not systematically distinguish between different types of research and learning activities [25,26]. Therefore, we searched for criteria or a particular logic that would provide some kind of categorization of the different types of SeConts. Some criteria that provide useful approaches to classification in other biomedical contexts are inappropriate. For example, although categorization by medical specialty seems conceivable at first glance, we determined it to be unhelpful on closer inspection. With all the different specialties that exist in medicine, such a categorization does not provide any reduction of heterogeneity. Categorization into common typologies of biomedicine, such as basic research, clinical research, and translational research, is also not useful, because learning activities, as described above, cannot be clearly assigned here. Ultimately, we found a categorization in terms of the object of investigation (ie, the entity about which SeConts seeks to produce knowledge) using three levels most appropriate. These three levels are familiar from the social sciences: micro level, meso level, and macro level. The object of investigation can be either patients (micro level), clinical care units such as clinical departments or single hospitals—a specific delimited patient population (meso level), or the general public (macro level). Taking the object of investigation as a criterion, we can attribute each type of research or learning activities to a general area of application: non-interventional (observational) clinical research focusing on individuals, quality control and improvement (and similar uses contributing to the creation of an LHCS) focusing on clinical units, and public health research focusing on the general public. Table 1 sums up all types of the aforementioned research or learning activities, including the object of investigation and area of application. It also shows that some types of research or learning activities are not limited to a single object of investigation or area of application.

After introducing the individual factors that determine the probability of unauthorized reidentification and the magnitude of harm implied by SeConts, questions arise concerning their application and use. How do these factors form a comprehensive framework for risk assessment in the context of SeConts? How can the framework be applied? How can the assessment results be used? Who may apply and use this framework?

With our risk assessment, we intend to support the analysis, evaluation, and potential decision-making process of research ethics committees and data use and access committees, as well as scientists, bioethicists, and funders investigating the ethical acceptability of requests for concrete types of SeConts. We recommend the following procedure with five consecutive steps to apply and use the risk assessment framework.

Having presented our risk assessment, its application and use, and illustrating its applicability by means of a concrete example, we will address the limitations and possible criticisms of our framework. Two objections can be raised against the risk assessment approach. First, we classify a priori certain data types according to their harm potential as a basis for operationalizing the magnitude of harm. However, it can be argued that such an a priori classification does not take into account the information that can be inferred from the data beyond the apparent information content [72,73]: Information that is considered potentially harmful (eg, sexual orientation, religious beliefs) can be derived from information that would probably be considered harmless a priori (eg, Facebook likes) [74]. We are aware that our a priori classification of data does not consider possible inferences that can be drawn from data in the future. However, there is an important difference in whether data contain information about a certain diagnosis or whether this diagnosis can only be assumed with a certain probability. Likewise, possible inferences that can be drawn from the data can rarely be predicted a priori. Therefore, our approach is limited to identifying the types of data that have a clear potential to cause harm themselves.

A second possible criticism of the proposed approach is conceivable. Regardless of whether information is inferred indirectly from data or whether the information is contained directly in it, the same data can entail different levels of individual risk for different people. Risk assessment, such as the one we present here, cannot reflect these different levels of individual risk. Both the probability of unauthorized reidentification and the possible magnitude of harm can be very different for the same data types in different people. The x-ray of a patellar luxation may be considered as nonsensitive information for most people, especially as it does not contain any identifying information. Nevertheless, a professional soccer player might disagree, as the information in the data contains the risk of unauthorized reidentification (due to possible analog and comparable x-rays of his club) and could be potentially harmful to his career (eg, through discrimination in relation to a possible contract extension). Our approach cannot depict these individually possible risks and can only provide guidelines for the assessment of generally expected risks. The risks of individual persons with special risk profiles must be addressed using individual measures. Among other things, this calls for the establishment of a suitable information and consent procedure or an opt-out option that allows persons with an individually high risk to decide for themselves whether this risk is too high or not. Which model of informed consent may be appropriate is not the subject of this study.

In this paper, we addressed three desiderata of the current literature on SeConts. In the first step, we clarified the concept of SeConts. To this end, we analyzed each element of the concept and then provided a comprehensive definition of SeConts as activities that solely use data produced for the sake of health care and in the context of health care to improve biomedical science or services.

In the second step, we illustrated the scope and practical relevance of SeConts by providing a list of concrete types of research or learning activities that can be subsumed under the concept. These types of research or learning activities were roughly classified as either non-interventional (observational) clinical research, quality control and improvement, or public health research.

In the third step, we provided a framework for risk assessment for SeConts, focusing on the risks for patients related to informational self-determination. By operationalizing the concept of risk for application to SeConts, we identified factors that determine the probability of unauthorized reidentification as well as the magnitude of harm of a potential harming event implied in SeConts. We then discussed the application and use of our risk assessment framework and presented a practical example of a concrete study to illustrate its application.

Through our conceptual clarification of SeConts, we created a basis for understanding what SeConts means. The analysis of its scope shows that SeConts can realize its potential in a broad field of medical research. This illustrates the high practical relevance of SeConts. The risk assessment presented can be applied as an essential building block for an ethical evaluation of concrete SeConts conducted by research ethics committees and data use and access committees, as well as scientists, bioethicists, and funders. It can thus benefit the safe secondary use of clinical data in data-gathering, non-interventional research or learning activities.