Hitting the (bio)mark Part 2: analysing, interpreting, and reporting biomarker data in cardiovascular research

Abstract

Incorporating biomarkers into cardiovascular studies, including nursing research, is a common approach when identifying underlying mechanisms and providing targets for intervention. However, effective utilization of biomarker data demands careful consideration. In the analysis, interpretation, and reporting phase, there are many facets to consider, including non-normality of the data, normalization procedures, and potential confounding influences of other clinical data. Furthermore, as many studies focus on patient-reported outcomes (PROs), it is important that the analysis and interpretation of biomarkers in relation to PROs is rigorous and reproducible. In this article, Part 2 of 2, we provide an overview of considerations for the analysis, interpretation, and reporting phases of biomarker studies. We also provide an example of these steps.

Introduction

Incorporating biomarkers into cardiovascular nursing research is a common approach when identifying underlying mechanisms and providing targets for intervention. Moreover, the inclusion of biomarkers into nursing research studies may help advance the movement towards precision health related to patient-reported outcomes (PROs).¹ The inclusion of biomarkers can be as simple as a single plasma protein [e.g. troponin or N-terminal pro-B-type natriuretic peptide (NT-proBNP)] that is measured in a clinical core lab, or as complex as running multiple assays for various genetic, transcriptomic, metabolomic, or proteomic markers. As described in Part 1, many considerations are needed to ensure a rigorous and thorough methodology for identifying a biomarker (or multiple biomarkers), designing experiments, selecting an assay, and processing samples. After the biomarker data are resulted, there are additional considerations in analysing, interpreting, and reporting the data, particularly in the context of other clinical data. Furthermore, as many cardiovascular nursing research studies focus on PROs, it is important the analysis and interpretation of biomarkers in relation to PROs are rigorous and reproducible. In this second part, we describe how to analyse, interpret, and report biomarker data that are resulted from sample processing (Central Illustration; Box 1 for common definitions; Table 1 for examples). We also present an example to illustrate the steps and considerations.

Step 5: perform the initial analysis and report

How a biomarker is initially analysed and reported will depend on a number of factors, including the biological material from which it was derived (e.g. plasma or tissue), the type of assay that was used, and any modifications needed after raw values are obtained (e.g. normalization, log transformation). First, there are several pre-analytic steps to produce the absolute value of the biomarker, depending on the experimental procedure. Oftentimes this involves using a standard curve to quantify the biomarker. For example, with an enzyme-linked immunosorbent assay (ELISA), a standard curve is generated based on standards of known concentrations tested alongside the samples. The standard curve then allows you to quantify the concentration of each sample based on readouts from the plate reader. At this point, you have ‘raw values’ of biomarkers. Biomarkers are typically presented as a certain amount per volume (fluid) or mass (tissue), also called a concentration (e.g. pg/mL). It is critical that the units are always reported alongside the biomarker values.

After obtaining the raw values of the biomarker, the next step is to perform simple descriptive statistics to understand range and variability, including any outliers. In order to minimize any potential confounding variability introduced across multiple assays, sometimes it is necessary to normalize the biomarker value across all assays. Normalizing across assays is possible when quality controls are included across assays (e.g. including the same pooled plasma across assays). One simple approach to normalization is to divide the biomarker value by the average of the quality control value in that assay. Of note, some assays, including the newer multiplex assays and mRNA sequencing, require normalization procedures to produce relative (and not absolute) values of biomarkers (e.g. normalized protein expression for multiplex assays⁶). Also, when quality controls are used across assays, intra- and inter-assay coefficients of variability can be calculated (see Box 1 for definition and calculations). These are typically reported in the methods section or supplementary material.

Oftentimes biomarker data are severely skewed, necessitating additional data manipulation, such as log transformation, to achieve normally distributed data suitable for common regression techniques. Skewness can be evaluated using visualization with histograms and skewness tests available in many common statistical software programs. Additionally, it may be helpful to create ratios of two related biomarkers (e.g. blood urea nitrogen-to-creatinine ratio) or calculate additional equations [e.g. homeostatic model assessment for insulin resistance (HOMA-IR) calculation⁷] to better understand the biological process. Throughout all of these steps, it is imperative to have rigorous and transparent methodology to ensure results are reliable and reproducible. Key steps include retaining all raw data files, tracking analytic steps and decisions, and having two investigators separately analyse the data to compare and validate findings.

Step 6: analyse biomarker data in relation to patient-reported outcomes

While biomarker data in and of themselves are interesting data points, and may in fact be the final step of analysis, understanding the biomarker data in the context of clinical information is very helpful in translating the findings. Furthermore, many readers may be able to understand common clinical biomarkers (e.g. troponin), but it may be difficult when novel biomarkers are measured; thus, it is important to answer the question ‘What do the data mean?’ Moreover, analysing biomarker data, common or novel, in the context of clinical information helps advance science towards precision health.

First, biomarker data (both raw and transformed data) are merged with other available data such as sociodemographic and clinical characteristics, PRO data, and so on. The type of analysis will depend on the research question. Common cross-sectional techniques involve generalized linear modelling techniques (e.g. linear or logistic regression, mediation analysis, moderation analysis). When many biomarkers are measured, especially those from the same or similar pathways (e.g. inflammation), multiple regression models may need to be evaluated to generate parsimonious models. For example, Kazukauskiene et al.⁸ examined associations between biomarkers of inflammation, thyroid function, and NT-proBNP (all natural log transformed) and health-related quality of life (HRQOL) among patients with coronary heart disease and heart failure. They first used univariate regression models to identify significant associations and then multivariate models including those significant in bivariate testing and adjusting for clinical characteristics.

Notably, there are multiple approaches within the family of generalized linear models that extend linear and logistic regression to data that have distributions other than normal.⁹ If longitudinal data are available, multiple techniques can be used such as mixed effects and latent growth curve modelling. There are many benefits to using latent growth curve modelling to analyse biomarker data, such as the ability to accommodate non-normally distributed data, missing data, and unevenly spaced time points (see example in Denfeld et al.⁴). Biomarker data can also be incorporated into mixture modelling approaches, such as latent class¹⁰ and growth mixture modelling,¹¹ as well as survival analysis¹² (see example in Willinger et al.¹³). Another consideration is the benefits and drawbacks to analysing biomarker data within cross-sectional vs. longitudinal designs. While cross-sectional biomarker studies can provide insight into associations between biomarkers (especially novel biomarkers) and outcomes of interest, they are limited in their ability to provide insights into the natural progression of a biomarker, its variability, and its relationship to disease development or progression. However, longitudinal data are needed in biomarker research due to their ability to capture patterns, trends, and individual variations over time, offering valuable insights into disease progression, treatment effectiveness, and personalized healthcare.

Step 7: consider other important covariates

When working with human models of disease, especially involving PROs, it is necessary to consider and incorporate other factors that may influence or confound the relationship being examined. When examining the association between a biomarker and PROs, it is likely necessary to adjust for other relevant clinical characteristics that may influence the relationship; for example, using a composite score of relevant clinical variables when examining associations between biomarkers and symptoms¹⁴ or frailty.¹⁵ Furthermore, many biomarker research questions focus on comparing two known groups, such as males vs. females.^4,16 While simply comparing groups is one approach, a common criticism is not considering confounding variables that would explain the group differences. Similar to above, one approach is to include these confounding variables (single or a composite) in a regression model; however, with small sample sizes (a common occurrence in biomarker studies), you are limited in the number of variables that can be entered into a regression model. One way to approach this is through matching participants on key characteristics (e.g. age, sex, severity of disease) and then comparing biomarker values. The matching procedures can be as simple as matching on age or as complex as generating propensity scores¹⁷ for each individual and then matching on scores.

One important consideration when analysing biomarker data is related to sex as a biological variable (SABV). It is important to understand SABV from multiple angles, such as understanding if the biomarker is fundamentally different between males and females either physiologically or pathophysiologically (i.e. sexual dimorphisms) or if a certain value of the biomarker indicates different processes or manifestations between sexes. While there are a number of ways to approach SABV, the ‘must do’s’ include (i) including both sexes in the study unless there is a strong rationale not to include one sex (rare occasions), (ii) collecting relevant data related to SABV such as hormonal therapies or cycle phase, and (iii) communicating sex differences and similarities.¹⁸

Step 8: interpret and present the final results

The amount of time it takes to interpret the final results should not be underestimated, and this is where it is important to include collaborators. A first question to ask is: ‘what does the association mean?’ Sometimes the associations are intuitive at first glance; for example, higher beta-adrenergic receptor kinase-1, an enzyme that downregulates and desensitizes beta-adrenergic receptors, is associated with worse physical heart failure symptoms.¹⁴ But sometimes the associations are counterintuitive; for example, lower myostatin, a skeletal muscle cytokine, is associated with higher levels of physical frailty.¹⁵ With either scenario, it is critical to be well-versed in the literature to tease apart potential explanations for your findings. One key question to ask when reviewing the literature in your topic area is how did they measure, analyse, and report the biomarkers, and how is this similar to or different from your biomarker findings?

The presentation of the final results depends on the research question and can be tailored to the target audience using tables and figures. For biomarker data, you can consider presenting both raw and transformed data in order to have both clinical applicability and analytic transparency. Graphs of the data are very helpful in displaying the findings overall or by group, either at one time point or over time. For effective scientific data presentation using figures, prioritize clarity and simplicity: (i) use clear labels and titles, employing legible fonts and appropriate sizing; (ii) ensure axes are labelled with units and have clear scales; (iii) match the right type of graph (e.g. bar for group comparisons, scatter plot for linear associations) with the data; (iv) minimize clutter by avoiding unnecessary elements and emphasizing key findings using colours or annotations; (v) maintain consistency in style and formatting across figures for coherence; and (iv) provide detailed legends and captions to aid understanding. Remember, the goal is to convey complex information succinctly, fostering comprehension, and engagement for your audience.

In tandem with the presentation of the findings, the methods will need to be described in sufficient detail, including biomarker selection rationale, sample collection, processing (including type of assay used), and analytic steps. As word count limitations may influence the description, we suggest citing method papers and utilizing supplemental materials to ensure clarity and transparency.

Example application

To illustrate the analysis and interpretation of biomarker data in relation to PROs, we utilized a dataset from a previous study¹⁵ and ask the research question: ‘Is plasma glucose, insulin, or insulin resistance (i.e. HOMA-IR) associated with HRQOL among patients with heart failure?’. We assayed both fasting glucose and insulin concentrations in the plasma of patients with heart failure. As previously described, glucose was measured using a colorimetric assay (BioAssay Systems; Hayward, CA, USA), and insulin was measured using an ELISA (Mercodia; Uppsala, Sweden). To generate HOMA-IR, we used the equation: [fasting glucose (mmol/L) × fasting insulin (μU/mL)/22.5].⁷

The sample has been previously described.¹⁵ In the initial examination of biomarker data, insulin concentrations ranged from 1.5 to 66.0 mU/L [median 9.4 (interquartile range 5.0–16.7)] with significant skewness (Figure 1A). Glucose concentrations ranged from 71.3 to 319 mg/dL [median 105.1 (interquartile range 95–126)] also with significant skewness (Figure 1B). Thus, insulin and glucose were natural log transformed to generate more normal distributions (Figure 1C and D). HOMA-IR was calculated (after converting glucose from mg/dL to mmol/L), and this ranged from 0.32 to 30.60 [median 2.39 (interquartile range 1.30–5.48)]. Generalized linear modelling was used to analyse the association between the biomarkers and Minnesota Living with Heart Failure Questionnaire (MLHFQ) scores: total, physical, and emotional scale.

After adjusting for Seattle Heart Failure Model Score and Charlson Comorbidity Index (Table 2), the natural log of insulin was significantly associated with both total and emotional MLHFQ scores; that is, higher plasma insulin was associated with worse total and emotional HRQOL. Similarly, HOMA-IR was significantly associated with emotional MLHFQ scores; that is higher insulin resistance was associated with worse emotional HRQOL. There were no significant associations between the natural log of glucose and MLHFQ scores. These findings may indicate that worse HRQOL in HF, especially emotional HRQOL, may be associated with insulin resistance or related pathways (Figure 2). However, these are preliminary findings, and further analyses would need to investigate and pinpoint specific mechanisms.

When not to use a biomarker and alternative approaches

The decision to use a biomarker should be made carefully. For example, some biomarkers may be expensive to measure, and the cost-benefit ratio should be evaluated. Additionally, the decision to incorporate a biomarker into a study should be substantiated by ample evidence, with due regard to ethical implications. Moreover, when collecting samples in non-routine practice, the burden on participants must be considered. This involves being mindful of the potential impact on individuals contributing to research. Lastly, it is imperative to acknowledge that research, particularly involving biomarkers, results in significant waste. Therefore, a comprehensive examination of the consequences of biomarker usage on both the environment and the economy is essential.

Various alternative approaches exist for utilizing biomarkers in research without utilizing invasive procedures like venipuncture for sample collection. One option is to obtain stored samples from prior or ongoing studies, either from a collaborator or a biobank, for secondary analyses. Additionally, biospecimens can be collected from clinical waste products, such as discarded surgical tissue [e.g. cardiac tissue plug from left ventricular assist device (LVAD) surgery] or fluid from drains. Another avenue involves conducting secondary analyses using laboratory values from electronic medical records to explore the relationship between these biomarkers and patient outcomes. These methods offer valuable insights into biomarkers and their association with clinical or PROs. Given their often lower costs, these approaches are excellent for generating pilot data that can inform larger biomarker studies.

Conclusions

This Part 2 article emphasizes the important role of biomarkers in cardiovascular nursing research, offering a detailed roadmap for their incorporation and analysis. This paper, Part 2 of 2, focuses on crucial aspects of analysing biomarker data in relation to PROs, highlighting normalization and data manipulation. Furthermore, different statistical techniques and considerations for covariates were discussed to enable the translation of biomarker data into a clinical context. Lastly, it is essential to understand the literature and suggest potential explanations for your findings, while transparently presenting the results in figures or tables to facilitate comprehension. The two papers provide a comprehensive guide to equip researchers with a systematic approach, fostering precision health insights and enhancing the understanding of patient outcomes in cardiovascular nursing research.

Funding

This work is supported, in part, by Research Foundation Flanders (G072022N) to B.D., the National Institute of Nursing Research of the National Institutes of Health (R01NR019054) to Q.D., the Office of Research on Women’s Health and the Eunice Kennedy Shriver National Institute of Child Health & Human Development of the National Institutes of Health (K12HD043488) to Q.D., and the National Institute on Aging of the National Institutes of Health (K23AG076977) to B.B. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Data availability

The data underlying this article will be shared on reasonable request to the corresponding author.

References

Author notes