Questions & explanations
1. Compare how Rv changes when a vaccine reduces susceptibility versus when it reduces infectiousness, assuming the same efficacy and coverage. Which one leads to faster epidemic decline?
Both reduce Rv, but they act differently. A susceptibility-reducing vaccine protects the vaccinated from getting infected, directly reducing the pool of susceptibles. An infectiousness-reducing vaccine allows vaccinated individuals to get infected but spread less. For the same efficacy (say 50%) and coverage (60%), Rv with susceptibility reduction: 3 * (0.6*0.5 + 0.4) = 2.1. With infectiousness reduction: 3 * (0.6*0.5 + 0.4) = 2.1 as well, so numerically identical. However, susceptibility reduction prevents infections in vaccinated, while infectiousness reduction allows breakthrough infections but with lower spread. In practice, both contribute similarly to reducing Rv, but susceptibility reduction may lead to faster decline because it blocks infections entirely.
2. What is confounding in a clinical study?
Confounding happens when a third factor is linked to both the exposure and the outcome, making it seem like the exposure causes the outcome when it might not. For example, in a study on coffee and heart disease, age could be a confounder because older people drink more coffee and also have higher heart disease risk. To control for confounding, researchers can use methods like stratification, where they analyze data within age groups separately, or multivariable regression, which adjusts for multiple factors at once. Confounding is different from effect modification, where the effect of the exposure changes across levels of another variable. Detecting confounding requires comparing crude and adjusted estimates; if they differ, confounding may be present.
3. Give an example of a privacy concern in surveillance that is addressed by law.
A common privacy concern is that health data might be sold or used for purposes other than public health. For example, an insurance company might want to know who has a chronic disease to raise their premiums. Many countries have laws like the Health Insurance Portability and Accountability Act (HIPAA) in the United States or the General Data Protection Regulation (GDPR) in Europe that forbid using health data for such purposes without consent. These laws require that data be anonymized—meaning names and identifying details are removed—before it is shared. They also give people the right to see their own data and ask for corrections. By law, surveillance data must be used only for protecting health, not for discrimination.
4. Give an example of how the surveillance system in India differs from that in the United States.
India's surveillance system, called the Integrated Disease Surveillance Programme (IDSP), uses a mix of weekly reports from local health workers and labs to watch for 20 diseases like malaria and typhoid. In contrast, the United States has the National Notifiable Diseases Surveillance System (NNDSS), where each state reports cases of over 120 diseases to the federal Centers for Disease Control and Prevention (CDC). India relies more on paper-based reporting and community health workers, while the US uses electronic systems and automated lab data. Both aim to detect outbreaks, but India's system faces challenges with many remote areas and limited internet, while the US has more resources for high-tech tracking.
5. Compare disease surveillance in a low-resource setting with that in a high-resource setting.
In a high-resource setting like a rich country, surveillance uses computers, automated lab reports, and many trained staff to track diseases in real time. In a low-resource setting, surveillance often relies on paper forms, few health workers, and basic tools like thermometers. For example, a rich country can test every suspected flu case and see the virus type within hours, while a poor country might only test a few samples and wait weeks for results. Both aim to find outbreaks, but the low-resource setting has to prioritize which diseases to watch and use simpler methods. The rich country can afford to be more thorough, but the poor country adapts with community-based reporting and mobile technology.
6. Compare the legal approach to privacy in surveillance between a country with strong data protection laws and one with weak laws.
A country with strong data protection laws, like Germany, requires explicit consent from people before collecting their health data and imposes heavy fines for misuse. A country with weak laws, like some developing nations, may have no clear rules, so health authorities can collect and share data without much oversight. In Germany, a person can sue if their data is leaked, while in a weak-law country, there may be no legal recourse. Strong laws protect privacy but can slow down surveillance because of paperwork. Weak laws allow faster data collection but risk harming individuals. Both approaches have trade-offs, and many countries are working to find a balance that protects both health and rights.
7. Compare the impact of selection bias and information bias on study results.
Both selection bias and information bias can lead to incorrect estimates of the association between exposure and outcome, but they arise from different sources. Selection bias affects the composition of study groups, potentially making them not comparable. Information bias affects the accuracy of measurements, leading to misclassification. For example, selection bias might make a treatment group appear healthier because sicker patients dropped out, while information bias might misclassify some treated patients as untreated. Both can overestimate or underestimate the true effect. The direction of bias depends on the specific situation. Careful study design and conduct are needed to minimize both.
8. How is evidence graded in the GRADE system?
The GRADE system (Grading of Recommendations Assessment, Development and Evaluation) rates the quality of evidence from high to very low. High quality means further research is very unlikely to change our confidence in the estimate of effect. Evidence from randomized trials starts as high, but can be downgraded due to limitations like risk of bias, inconsistency, indirectness, imprecision, or publication bias. Observational studies start as low quality, but can be upgraded if the effect is large or dose-response is present. The final grade determines the strength of recommendations: strong or weak. This system helps clinicians understand how confident they can be in the evidence.
9. Give an example where compositional data analysis reveals a different finding than analyzing each food group in absolute grams.
Suppose a study looks at absolute intake of vegetables and meat in grams. It might find that higher vegetable intake is linked to lower disease risk, and higher meat intake to higher risk. But compositional analysis could show that the ratio of vegetables to meat is more important. For instance, if someone eats more vegetables but also more meat (keeping the ratio same), the absolute analysis might suggest benefit from vegetables, but compositional analysis would show no change in risk because the proportion didn't change. This highlights that swapping meat for vegetables is key, not just adding vegetables. Compositional analysis thus gives more precise dietary guidance.
10. Compare the surveillance system of a high-income country like Japan with that of a low-income country like Uganda.
Japan's surveillance system uses advanced technology, with hospitals sending electronic reports daily to a national database that tracks diseases like flu and COVID-19 in real time. Uganda's system, called the Integrated Disease Surveillance and Response (IDSR), relies on health workers in rural clinics filling out paper forms each week, which are then entered into a computer system. Japan can quickly analyze data and respond to outbreaks, while Uganda often faces delays due to poor roads and lack of computers. Both systems focus on the same goal—finding outbreaks early—but Japan has more money and staff, while Uganda adapts with simple tools and community involvement.
11. How can you control for confounding in a study?
There are several ways to control for confounding. In the study design phase, you can use randomization, which balances confounders between groups, or restriction, where you only include people with a certain level of the confounder. Matching is another method, where you pair participants with similar confounder values. During analysis, you can use stratification, which examines the exposure-outcome relationship within each level of the confounder, or multivariable regression, which statistically adjusts for multiple confounders at once. The best method depends on the study type and available data. Proper control helps get a more accurate estimate of the true effect.
12. How does G-computation differ from standard regression adjustment?
Standard regression adjustment includes confounders as covariates in a model for the outcome. However, if the model is nonlinear or there are interactions, the coefficient for treatment may not represent a population-average effect. G-computation explicitly standardizes the confounder distribution by predicting outcomes for everyone under each treatment level, then averaging. This gives a marginal effect that is interpretable even with interactions or nonlinear models. For example, with a logistic outcome, the odds ratio from regression is conditional, while G-computation yields a marginal odds ratio. G-computation also handles time-varying exposures more naturally.