91黑料

  1. Home
  2. Insights
  3. Blog
  4. SDoH data analysis for proactive outcome improvement: A multilayered approach

SDOH data analysis for proactive outcome improvement: A multilayered approach

Social determinants of health (SDOH) affect patients in widely variable ways, and undoubtedly have significant impact on patient health outcomes. Layers of unmet social needs (including housing instability, lack of education or poor access to transportation, among many other factors) exacerbate disparities in health outcomes that are already impacted by geography, race and ethnicity, age, ability, etc.i These SDOH are not binary, and they can impact different groups of patients in different ways. Proper analysis is required to gain actionable insights that can guide patient-centred decision making and enhance efficiencies throughout clinical development and commercialisation. 

Here we outline the multi-step, layered methodology to SDOH data analysis that enables closer examination of patients’ healthcare consumption and behaviour to optimise brand strategy and deliver better solutions and services to patients to achieve better outcomes.

Intelligent drivers analysis

One of the ways SDOH data can yield beneficial insights is through drivers analysis. Advancing technologies like artificial intelligence (AI) and machine learning (ML) are increasingly being used for the initial stage of this analysis as they can ingest the wide range of relevant data and arrive at relevant conclusions more expediently than human computation alone. SDOH data is typically layered on top of other datasets, like claims, lab or third-party data, to provide deeper analysis, and the sheer volume of data requires intelligent tools to properly integrate and manage. 

Refined ML algorithms can identify persistency drivers for patients, the elements that make them more or less likely to remain on their medications as prescribed. These drivers are refined through analysis of factors such as propensity towards adherence, ability to pay for medications, and health equity scores from SDOH models. 

Another key element in this first round analysis is dimensionality reduction. ML helps us to quickly determine the key drivers among a large array of input variables inherent to robust datasets, allowing analysts to hone outputs to exclude noisy data or irrelevant information that may skew results. We utilise statistical techniques such as Principal Component Analysis in combination with domain expert knowledge to focus on most impactful determinants and identify individual factors that are driving a patient’s persistency either in adherence to protocol or a product. For example, in a diabetes study, patient profiles may also have high incidence of hypertension which would be picked up by the algorithms as a significant driver but is ultimately irrelevant to persistency. 

Clustering patient profiles

The next phase of SDOH data analysis uncovers trends in patient profiles. This involves selecting and applying clustering techniques including k-means, DBSCAN, Gaussian mixture model and hierarchical clustering. Algorithms are trained using curated, relevant data to group patients by similarities in their social determinants and apply association rule mining to identify frequent co-occurrences of determinants. As a result, we establish unique clusters or cohorts of patients based on data proximity. 

Before these models can be used for insight generation, they must be evaluated using standard clustering validation metrics, e.g., Silhouette score and domain related metrics, comorbidity prevalence within each cluster, to fine-tune the model parameters and identify the optimum clustering approach. Data analysts will further review the models to scrub any noise from the results and ensure meaningful metrics are represented. The resulting models highlight natural groupings or associations within the social determinants that can indicate their influence on patient journeys and health outcomes without relying on pre-labelled data.

Profiling patient persistency

The positive and negative drivers are identified and clustered then ranked based on SDOH features with the input from human expertise. Understanding the profiles of patients within certain driver clusters facilitates proactive approaches to patient persistence instead of reactive responses. This exercise can identify trends in the potential drivers in relation to specific features of interest. For example, if a sponsor wanted to run a decentralised clinical trial, a key SDOH factor would be attitudes and acceptance of technology. Layering patient metrics, target population demographics and geography with attitudes toward technology can steer recruitment strategies for more successful engagement, adherence and persistence. 

SDOH insight generation

The rankings and results generated through drivers analysis and clustering are not yet actionable insights – they must be properly contextualised to extract meaning. This interpretation involves understanding the composition of the patient clusters, assessing the strength and implication of association rules, and potentially utilising visualisation techniques within business intelligence tools to better comprehend the inherently complex relationships between data points. 

Internal statistical measures validate the findings and consultation with healthcare experts ensures reliability of the results. The conclusions drawn balance ML and human expertise to provide a strategic understanding of the key social determinants of health that affect the patient population. These insights then enable more targeted interventions including outreach and recruitment, patient-centric trial designs, protocol adherence and patient support services during clinical trials or persistency in commercialisation stages. Meanwhile, our technologies allow us to take multiple tokens and match them to the patient master to connect deidentified patient insights across development continuum.

Proactive, iterative processes for better outcomes

Taking a proactive approach to identifying and addressing SDOH ensures patients have access to the treatment they need and are appropriately supported to adhere to it. The insights uncovered also support many aspects of brand management including forecasting, promotion strategy and tactics, patient support programs and even healthcare provider program development. As we continue to iterate and refine the processes we use to integrate, analyse and interpret SDOH data, we will gain deeper understanding of how, where and why the environment and social structures influence patient outcomes, helping to drive decisions supporting health equity.

Connect with us to learn more about leveraging healthcare to improve your clinical development strategy and patient outcomes.

i U.S. Playbook to Address Social Determinants of Health. November 2023

In this section

Connect with us