Diagnostic study
What are clinical diagnostic studies?
Planning diagnostic studies
Diagnostic studies are used to examine, evaluate and evaluate diagnostic procedures in medicine, the diagnostic accuracy, usually expressed as sensitivity and specificity, being the subject of interest. Currently, the effects and consequences of the diagnosis are also coming into focus: What is the effectiveness of a therapy selected based on the diagnostic test? The cost bearers in particular have this aspect in mind.
Diagnostic studies are a particularly interesting and challenging type of clinical study for two reasons. This is surprising at first, as it is ultimately only about the evaluation of a simple 2x2 table in which two true states of the patient (reference standard: sick: D , not sick: D-) two results of the diagnostic test (test result: test positive: T , test negative : T-) are compared. That the design and analysis of clinical diagnostic studies has to be more complex becomes clear when one looks at the checklist of the STARD statements
(Standards for the Reporting of Diagnostics Accuracy studies). Although actually intended for publication, it is helpful to consider the checklist when planning. This helps to set the right study objectives and the right study population, avoid bias, address measurement quality appropriately, and draw the right conclusions.
The interesting thing about this type of study is the conceptional work. It is important to (i) correctly represent the intended use, (ii) the clinical application situation and (iii) the location in the diagnostic path. This is more difficult than expected; especially when developing new diagnostics, it can be seen that these aspects were sometimes neglected during development.
Diagnostic studies are divided into different phases. Here the phase division according to Köbberling et al. very proven, see this link.
Bias in diagnostic studies
A second important point are the biases that threaten when conducting and analyzing diagnostic studies. The focus is on the spectrum bias. In the case of a case control design, ie the recruitment of patients based on their disease status, there is an unintentional exclusion of "unclear" cases, ie the inclusion of only "clear" cases, and thus underrepresentation of low disease stages and "gray zone" cases. In the case of those who are not ill, there will be a distortion towards healthier, younger people and patients / test subjects with no age-related comorbidities. As a consequence, the application situation is not mapped correctly and the diagnostic accuracy of the procedure is overestimated (!). In an extreme case, the non-disease group is chosen incorrectly: If the intention is to use a diagnostic tool to distinguish between two disease states (e.g. if there are symptoms to differentiate between tumor and inflammation), then the study should not examine "sick" vs. be made "healthy".
The insufficient addressing of the selection of suitable study populations and their composition is the reason for the failure of many biomarkers that initially appear to be promising. The way out is to conduct the study as a cohort study, the inclusion criterion is required diagnosis to be used when the disease status is unknown.
Additional biases are often associated with the reference standard, ie the definition of a patient's group membership (D , D-). The reference standard can e.g. themselve have only a limited diagnostic accuracy. Or it is determined for groups D and D- in different ways and, if necessary, with different quality.
ACOMED statistics specializes in the design, implementation and the analysis of diagnostic studies. Contact us if you need biometric support for the design and analysis.
