Occupational Risk Profiles for Infectious Disease Transmission in the Philippines: A Data-Driven Clustering Analysis

This study employs data-driven clustering of job-specific indicators to identify distinct occupational risk profiles for infectious disease transmission in the Philippines, revealing a socioeconomic gradient where lower-income roles face higher exposure risks and offering a framework for targeted public health interventions.

The Gist

Imagine the Philippines as a giant, bustling city where everyone has a different job. Some people work in quiet, air-conditioned offices with their own private rooms, while others work in crowded markets, busy hospitals, or on noisy construction sites.

This paper is like a detective story that tries to figure out which jobs are the most dangerous when it comes to catching infectious diseases (like the flu or tuberculosis). Instead of just guessing or grouping jobs by broad categories (like "healthcare" or "service"), the authors used a computer to look at the actual details of 986 different jobs and sort them into natural groups based on two main things:

1. How much you are exposed: How many people you see, how long you stand near them, and how crowded your workplace is.
2. How much money you make: This acts as a clue for how much power you have to protect yourself (like buying good masks, taking sick leave, or working from home).

The "Smart Sort" Experiment

The researchers tried five different ways to sort these jobs, kind of like trying different recipes to bake the perfect cake. They used methods called K-means, Hierarchical clustering, and others.

• The Winners: Two methods worked best. One (K-means) sorted the jobs into 4 big groups, and the other (Hierarchical) sorted them into 6 more detailed groups.
• The Losers: Some methods got confused or made groups that didn't make much sense, so the researchers decided to focus on the two winners.

The Four Main Groups (The "Big Picture")

Using the best method, the researchers found four distinct "neighborhoods" of jobs:

1. The "Frontline Fighters" (High Risk, Lower Pay):

   • Who: Doctors, nurses, police officers, teachers, and childcare workers.
   • The Situation: These people are constantly hugging, talking to, or helping strangers. They are the first to get sick because they can't easily avoid crowds.
   • The Catch: They often have lower incomes, which means they might not have the money or the job flexibility to stay home if they feel sick or to buy the best protection.

2. The "Office Professionals" (Medium Risk, Higher Pay):

   • Who: Engineers, scientists, managers, and IT specialists.
   • The Situation: They interact with people, but usually in controlled environments like offices or labs. They have some risk, but it's manageable.
   • The Advantage: They have higher incomes and better job security, so they can afford better safety gear or work from home if needed.

3. The "Mixed Bag" (Variable Risk, Lower Pay):

   • Who: Construction workers, drivers, retail staff, and service workers.
   • The Situation: This is a huge, messy group. Some days they are safe; other days they are in very crowded places. Their risk changes a lot depending on the specific task.
   • The Catch: Like the frontline fighters, they often have lower incomes, making it hard to control their environment.

4. The "VIPs" (Low Risk, High Pay):

   • Who: CEOs, judges, high-ranking military officers, and airline pilots.
   • The Situation: These jobs are usually done in very controlled, private, or regulated spaces. They rarely have to stand in a crowded line or touch strangers.
   • The Advantage: They have the most resources and the most control over their safety.

The "Six-Group" Zoom-In

When the researchers used the more detailed method (the 6-group version), they found even more interesting splits. For example, they realized that while all "doctors" are high-risk, the type of doctor matters. Some high-level medical experts are in a group with very high pay but still high risk, while support staff are in a group with lower pay and high risk. It showed that money doesn't always equal safety, but it does help you manage the risk better.

The Big Takeaway

The main lesson from this paper is that risk isn't spread evenly.

• The Unfairness: The people who are most likely to catch a disease are often the ones with the least money and the least power to change their situation.
• The Solution: The authors say we shouldn't just treat "all workers" the same. Instead, we need to look at these specific groups. If we know exactly which jobs are the "Frontline Fighters," we can give them the most protection, the best masks, and the most support, rather than trying to fix everything with a one-size-fits-all rule.

In short, the paper uses math to draw a map of the workplace, showing us exactly where the "danger zones" are and who is living in them, so we can help the right people first.

Technical Summary

Technical Summary: Occupational Risk Profiles for Infectious Disease Transmission in the Philippines

Problem Statement
Infectious disease transmission in workplace settings is driven by specific occupational activities that determine contact patterns, exposure duration, and environmental conditions. While existing literature acknowledges that occupations involving sustained close contact and crowded conditions face elevated risks, current assessments in the Philippines often rely on broad sector-based classifications or descriptive qualitative approaches. These methods fail to capture the granular variation in risk across specific job roles and largely overlook the interplay between exposure dynamics and socioeconomic vulnerability. Consequently, there is a lack of quantitative, data-driven frameworks that characterize infectious disease transmission risk across the Philippine workforce to inform targeted public health interventions.

Methodology
This study employs an unsupervised machine learning approach to profile occupational risk using a dataset of 986 occupations in the Philippines. The analysis integrates two primary variables:

1. Infectious Disease Spread Risk Score: Derived from the "Job Risk Calculator" (based on the framework by Dy and Rabajante, 2020), this score quantifies exposure based on four determinants: encounter rate (interactions per hour), work shift duration, crowd density, and protection level (PPE, hygiene, engineering controls). The score is computed as:
   $$Risk\ Score = \frac{Encounter\ Points + Work\ Shift\ Duration\ Points}{2} \times Crowd\ Density \times (1 - Protection\ Level)$$
2. Average Monthly Income: Used as a proxy for socioeconomic vulnerability and the capacity to adopt risk-mitigation measures.

The study applies and compares five unsupervised clustering algorithms to group occupations based on these two dimensions:

• K-means clustering (partition-based)
• Hierarchical clustering (agglomerative, using Ward's linkage)
• Spectral clustering (graph-based)
• Gaussian Mixture Models (GMM) (model-based)
• DBSCAN (density-based)

Cluster validity and robustness are assessed using internal metrics: the Silhouette Score, Davies–Bouldin (DB) Index, and Calinski–Harabasz (CH) Index. The study prioritizes methods that demonstrate stable, interpretable, and well-separated groupings.

Key Results
The analysis reveals that occupational risk is not uniformly distributed but follows a distinct gradient linked to socioeconomic positioning.

• Methodological Performance: K-means ($k=4$) and Hierarchical clustering ($k=6$) produced the most stable and interpretable structures, evidenced by higher Silhouette and CH scores and lower DB indices. Spectral clustering, GMM, and DBSCAN yielded less coherent groupings or were overly sensitive to parameter selection.
• K-means Clustering ($k=4$): Identified four distinct profiles:
  • Cluster 1: High risk, low-to-moderate income. Dominated by healthcare (physicians, nurses), emergency response, caregiving, and high-contact service roles (teachers, transport).
  • Cluster 2: Moderate risk, higher income. Composed of professional, technical, and managerial roles (engineers, scientists, IT specialists) with structured environments.
  • Cluster 3: Variable risk, low-to-moderate income. A heterogeneous mix of skilled trades, logistics, and service roles.
  • Cluster 4: Low risk, high income. Senior leadership, executive, and specialized roles (judges, pilots) operating in controlled environments.
• Hierarchical Clustering ($k=6$): Provided a more granular segmentation, further distinguishing within the high-risk and moderate-risk categories. Notably, it separated high-ranking executive roles (low risk, high income) from professional/technical roles (moderate risk, higher income) and identified a specific "core high-exposure" cluster (Cluster 6) dominated by clinical and emergency personnel with a wide income spectrum, distinct from the broader service-oriented high-risk groups.

Key Contributions

• Data-Driven Stratification: The study moves beyond broad sector classifications to provide a job-level risk stratification framework specific to the Philippine context.
• Integration of Socioeconomics: It explicitly demonstrates how infectious disease exposure risk correlates with income levels, highlighting that high-risk roles are often concentrated among workers with lower economic capacity to mitigate risks.
• Methodological Comparison: It evaluates the efficacy of multiple clustering algorithms for occupational health data, establishing that partition-based and hierarchical methods are superior for this specific dataset structure.

Significance and Claims
The authors claim that this study provides a practical framework for targeted workplace interventions and equitable allocation of public health resources. By identifying latent groupings that capture shared exposure characteristics and socioeconomic context, the findings support:

1. Risk-Informed Surveillance: Prioritizing high-risk occupations (e.g., healthcare, emergency services, and high-contact services) for monitoring and resource distribution.
2. Policy Planning: Moving away from one-size-fits-all sector policies toward occupation-specific strategies that address both exposure risks and structural vulnerabilities.
3. Equity: Highlighting the structural inequality where high-exposure workers often lack the economic resources for risk mitigation, suggesting a need for hazard compensation and strengthened safety standards.

The paper concludes that while the risk scores are based on generalized modeling and cross-sectional data, the integration of machine learning with public health principles offers a robust, evidence-based approach to addressing complex occupational health challenges in the Philippines.