From research cohorts to the patient – a role for “omics” in diagnostics and laboratory medicine?

Biomarkers in clinical research

It is important to distinguish the requirements for biomarker analyses in (bio-) medical research from the requirements in medical laboratory diagnostics. Typical relevant application for a “research use test” are in the field of drug development. Especially the question whether a drug candidate in a test system (cells, animal model) exerts any biological effect at all or not – as a screening tool for potential effects of newly synthetized compounds. Here, the change of readout peak patterns in a model system may be sufficient without specific identification of individual analytes. Analytics in the context of pharmaceutical development are regulated by the Food and Drug Administration (FDA) and the European Medicines Agency (EMA) and are completely independent of in vitro diagnostics (IVD).

In the context of clinical (biomarker) research, mass spectrometry (MS)-based multi-metabolite kits are now offered for decentralized use (for example by Biocrates); and service providers conduct metabolomics testing in corporate laboratories (Metabolon, Biocrates, Lipotype and others).

In terms of definitions, there is also a lack of clarity regarding the word “clinical”. Clinical research is sometimes understood merely as the opposite of in vitro investigations or animal studies, i.e., as soon as humans or human samples are the object of investigation (e.g., also in cohort studies). However, “kline” is borrowed from the Greek, meaning bed. Thus, in a narrow sense, clinical studies are those that deal with seriously ill patients.

Translation into diagnostics

The basic omics concept offers two different lines of translation into medical diagnostics: first, biomarker discovery starting from (multi-)omics procedures with the highest possible coverage aiming to identify single markers, often via retrospective or prospective cohort studies. Single markers identified in this way can then eventually be quantified in routine diagnostics using conventional platforms. For example, the goal of MS-based proteomics studies may be to identify single diagnostically valuable proteins that are subsequently translated into a diagnostic test in form of immunoassays, applicable to routine diagnostic high-throughput analyzers. A second concept of translation into diagnostics can in principle be to actually use analytical methods with a high analyte coverage in routine diagnostics, and to apply pattern recognition methods based on this [3]. In principle, this does not necessarily require the identification of individual analytes, e.g., eluting in one chromatographic peak. The risk of failure is high in both scenarios [4].

Unmet needs

Diagnostic gaps and unmet needs exist in today’s laboratory diagnostics especially in (i) tumor diseases with regard to early detection and prediction of individual biological behavior (especially prostate and breast carcinoma), and (ii) in the non-invasive diagnostics of atherosclerotic vascular diseases [6, 7]. These two disease areas contribute most to fatal outcomes and thus might benefit the greatest from improved diagnostics.

Multidimensionality and “probabilistic” results

Many physicians have the expectation that complex, polygenic diseases will require complex, multiparametric laboratory testing, rather than the classic single-marker approach, to improve accuracy of diagnosis. Here, a growing number of physicians have high hopes and anticipations for artificial intelligence [8] and machine learning [9], although few have a clear understanding what actually happens in these procedures.

Actionable results

The nuts and bolts of any laboratory parameter is the intended purpose (what is to be achieved) or intended use (how is the product to be used). Results of tests are usually only of medical value if they are actionable, i.e., if they can usefully inform concrete medical decisions. It is not a question of obtaining information as a value in itself, which can later possibly be used for scientific evaluation. This is the essential difference between clinical laboratory diagnostics and biomedical clinical research, where results should primarily be “interesting” but do not have to be directly “useful” and actionable.

One-dimensional and multi-dimensional laboratory diagnostics

Nowadays, clinical chemistry laboratory tests are generally “one-dimensional” in nature. This means that a single laboratory parameter (e.g., serum creatinine concentration) is interpreted in relation to its assigned biological reference range or medical decision limit, and individual decisions are made based on these. Examples include the initiation of differentiated nephrological diagnostics for a serum creatinine value above a decision limit, e.g., 110 μmol/L in a young man. Or the question of vitamin B12 deficiency, triggering the testing for methylmalonic acid in blood or urine, with clearly elevated values proving deficiency and resulting in initiation of a therapeutic intervention.

However, examination procedures which evaluate patterns of single values in the synopsis are hardly clinically established. In some areas, ratios or similar are calculated, e.g., the ratio of aldosterone to renin in the diagnosis of secondary hypertension, or the phenylalanine to tyrosine ratio in newborn screening for phenylketonuria. Scores are calculated from individual variables in a certain number of indications, e.g., the Model for End-stage Liver Disease (MELD) score for prognosis assessment in liver disease (composite tests).

As a rule, a diagnostic process is to be seen as a mosaic, in which the physician evaluates a multitude of anamnestic, physical and technical findings in the synopsis from their professionally based intuition. But a pattern evaluation of a multi-dimensional biochemical profile is currently not applied as an essential part of a diagnostic in clinical chemistry diagnostics. Clearly, the technological advances of multi-omics technologies have been an enabler in the worldwide growing attempts in the field of translational precision medicine, toward personalized healthcare based on molecular pathology.

Adoption

A key question is to what extent physicians would accept findings that primarily indicate probabilities (e.g., for the presence of a disease or its prognosis), generated from a large number of individual measurement data from analytes which may not be unambiguously identified or absolutely quantified. This is difficult to predict. The use of transcriptome analyses in risk stratification in breast cancer speaks at least to a possible acceptance of this approach by some physicians.

Technological aspects

From the analytical perspective, the two main challenges of omics-protocols are (i) the enormous diversity of target compounds with respect to physicochemical properties, and (ii) the huge dynamic range of their concentrations.

Correct identification of analytes (primarily in untargeted analyses) poses another great challenge, especially since many metabolites exist as structural isomers or isobaric compounds. Strikingly, the concept of metrological traceability as a fundamental principle in laboratory medicine quality assurance [11, 12], and its application to omics-based diagnostics still is an almost uncharted territory [13].

Specific risks and challenges of diagnostics for chronic diseases

The majority of healthcare resources in developed countries (but increasingly globally) are used in the context of chronic, non-communicable diseases (diabetes, cardiovascular disease, dementia, tumor disease, etc.). Demonstrating the benefit of novel laboratory tests in this area is particularly challenging, as studies of very long duration (in the order of decades) and in large patient collectives (in the order of thousands of individuals) are required to clearly evaluate a possible additional benefit of a new laboratory test. For example, the benefit of cardiovascular risk prediction (with consequent attempts of disease primary prevention) based on an innovative test in the normal middle-aged population can only be meaningfully assessed after about 20–30 years. Here, the risks (and perhaps ethical challenges) of such studies also become apparent: if a patient is assumed to be at low risk of developing cardiovascular disease based on such a new test, primary prevention of statins may not be started (possibly even despite known classical risk factors such as a family history, for example). If that patient eventually suffers a myocardial infarction at age 60, this may be seen as harm caused by the unreliable and “false-negative” risk test.

In stark contrast to chronic diseases, acute clinically relevant questions such as “is there a bacterial or a viral infection?” can be evaluated in very short periods of time depending on the natural history of a disease.

Test mathematics

Screening tests in the low-prevalence range should be the subject of particularly critical scrutiny, as potentially large numbers of false-positive results are generated, with a very low level of positive predictive value.

Effort and return – medical and economic

Omics procedures in patient-related medical laboratory diagnostics will require a great deal of effort for a market launch. Due to the complexity, which is disproportionately higher than for previous “one-dimensional” examination procedures, the effort for the development of such procedures up to a market launch will approach the dimensions required for the development of a novel pharmaceutical. On the other hand, the potential earnings prospects for diagnostics are generally much lower than for pharmaceuticals.

Studies with purely prognostic value are usually questionable. In the absence of an effective therapy option (for example the quantification of the risk of developing dementia or even a pre-symptomatic diagnosis) is problematic. The usefulness of laboratory tests is therefore fundamentally inseparable from the availability of treatment options in a specific situation, hence actionability.

Results of potential examinations without clear intended use, or without defined decision limits, could be understood “volatile” as an element in an intuitive overall picture of the physician of a patient. However, this contradicts the basic attitude of evidence-based medicine, which strives to place clinical decisions on a scientifically sound footing in a comprehensible manner [22, 23].

Once clinical validity and utility has been established, conceivable omics-based diagnostic services and products could take the form of i.e., examination of samples in one or several medical laboratories based on in-house procedures (i.e., “laboratory-developed tests”, LDT); analytical kits implemented and applied on general laboratory equipment in medical laboratories; or dedicated systems, i.e., special analytical systems with the dedicated purpose of the respective omics analysis [3].

Health technology assessment (HTA)

Compared to the formal requirements for placing an IVD on the market, the regulations for reimbursement by insurance companies represent a much higher hurdle overall. While the regulation of marketing essentially checks whether the intended use of an IVD is fulfilled and the product is safe, the HTA questions the actual medical added value of an IVD; specifically the additional benefit with regard to existing diagnostic procedures [24]. In Germany, this is the responsibility of the G-BA (Gemeinsamer Bundesausschuss) which carries out a scientifically very well-founded evaluation procedure. No resources from the statutory health insurance are granted for low-value tests. Such services can be offered individually by physicians, but then consequently only have a limited market potential, since the group of self-payers is rather small. National HTA agencies are established in many countries. In some countries, direct-to-customer (DTC) offerings are an essential business model, but ultimately only truly useful tests and high-value care will have lasting success. Fundamentally, DTC deals are problematic. Setting indications for complex laboratory diagnostic tests and, above all, interpreting them in an individual context is associated with major risks for the quality of life of the “customers” outside the scope of medical expertise and responsibilities. In particular, the use of risk metrics can deeply upset people, especially if the results are not interpreted and communicated by consulting physicians.

Overall, efforts of a utilization management are gaining importance worldwide. It is important to note that clinicians are generally not interested in additional data per se; ultimately, they are only interested in data that lead to measurably better treatment outcomes or cost saving. The subjective and emotional liking of data per se undoubtedly underlies much of the DTC business, but it is neither sustainable nor beneficial.

It remains to be said that clinical diagnostic laboratory tests must meet fundamentally different requirements compared to biomarkers in the field of biomedical research. The complexity of looking at individual patients is paradoxically much higher than when looking at more or less homogeneous cohorts in biomedical research, or even very homogeneous data e.g., from inbred animal models.