Epi in the News!

An interesting article from the New York Times, especially for those of you preparing for the final (Paper 2) exam.

As a member of the mental health group at LSHTM, I have an immediate answer to this question posed in the title. But as an epidemiologist, there's a lot in this article to consider.

For example, 

- What are the (unexpected) mediating effects of the trial interventions?

- What do you make of "prevalence inflation" ? How would you test this hypothesis?

- Explain the distinction between efficacy and effectiveness, using an example from a trial in the article

- Should we continue to fund trials into school-level interventions? If so, what needs to change in the trial design?

From The New York Times.

Over time officials identified four former clients and a sexual partner who had received H.I.V. diagnoses between 2018 and 2023, despite reporting few risks associated with infection, such as injection drug use, blood transfusion or sexual contact with a new partner.

1. As above, there was an infection outbreak and the transmission route was unclear. What steps should an epidemiologic take?

2. What study design is most suitable here? Why?

3. Assuming you could administer a questionnaire to the people who had been infected, how would you figure out what 'exposures' to ask about?

This article from the New York Times is terrific for considering confounding.

Using just the information in the article, can you construct a confounding triangle?

What does the triangle look like for the alternative explanation proposed by Elizabeth Cook?

Given these findings, do you feel that teen pregnancy is a causal factor for premature death? How would you design the next study to untangle this?

From The Guardian. This is a rare, welcome example of an article which presents the raw figures underlying the RR (the epidemiologic measure of effect). And so we can get start to consider the public health impact.

1. What is the PICO for this study?

2. Calculate the relative risk (RR).

A vaccine could reduce by 80% the numbers of babies and young children admitted to hospital with respiratory syncytial virus (RSV), a “groundbreaking” study has found.

3. How does the RR you calculated in #2 connect to the statement above?

4. Explain whether this RR is an efficacy or effectiveness figure.

5. Was it ethical to randomly allocate babies to a placebo?

The research found that, of the babies who received the vaccine, only 11 (0.3%) were hospitalised, in comparison with the 60 babies (1.5%) who were hospitalised after receiving just the standard care.

6. Are these incidence or prevalence figures?

7. Calculate and interpret a number needed to treat (more specifically, the number needed to vaccinate)?

In England, RSV is a leading cause of infant hospitalisation, with nearly 31,000 children aged four and under admitted each year with conditions linked to the virus. RSV causes between 20 and 30 infant deaths a year in the UK,

8. With 100% RSV vaccine uptake, what is the impact on these figures? (slight trick question, but I'm curious to read what you say!)

A long article from The New Yorker, which covers a wide range of ground of interest to epidemiologists. So, I'll put out only one question, though welcome your comments...

By the early sixties, Selikoff had collected data on a cohort of six hundred and thirty-two men who had worked in the insulation factory, some for many years. Among these men, Selikoff documented forty-five cases of lung cancer and mesothelioma—seven times more than the expected number. The incidence of stomach, colon, and rectal cancer was three times higher than expected.

1. What kind of analysis was used to get these findings? What is the epidemiologic measure corresponding to "seven times more" and "three times higher"?

From The Guardian

I found myself scratching my head after reading this. Perhaps one of you can figure out what is going on! But for me there are important details omitted in the article.

1. What are the PICOs?

2. Should the comparison arm of an HIV vaccine study be a placebo vaccine?

3. Why is it sometimes necessary to stop a trial early?

4. What is the statistical rationale of completing the study in Uganda, South Africa, and Tanzania, and not, say, in the US or Europe?

5. What is the rationale for using an RCT to evaluate an HIV vaccine, rather than another epidemiologic study design?

From the New York Times. This article is notable for being about an epidemiologist with a perceived conflict of interest.

1. Why do you think a trial was justified?

2. Should scientists be prohibited from managing studies when they already favour one outcome?

And an episode of the Milk Street podcast. The bit in question comes very early in the episode. (And here's a longer article from The Atlantic, where the journalist interviewed on Milk Street gets deeper into the epidemiological weeds.)

1. What is the RR figure to get 22%? What is the PICO (population, intervention group, comparison group, outcome).

2. What is the hypothesised mechanism?

3. What would you do if you did a study and had an unexpected finding like this?

From the Guardian

Here's an interesting article involving routinely collected data in England.

1. What is the study design?

2. What are the exposure and outcome?

2b. What is the mechanism of effect?

Overall there has been a drop of 3.7 admissions per 100,000 zero to 18-year-olds since 2018.

3. Is the 3.7 an incidence or prevalence figure? Is it a difference or a ratio?

4. Can you come up with (epidemiologic) alternative explanations for this drop?

5. Do we need a trial to prove causation? What would this involve?

Today's article is from The Guardian

Using data from the UK Biobank study, a long-term study tracking the health and genetics of adults across the UK, the authors looked at five different kinds of social connection reported by 458,146 people with an average age of 57 and then followed them for an average of 12.6 years.

1. What is the study design?

People who were never visited by friends or family were 53% more likely to die from cardiovascular disease and had a 39% increased risk of death compared with those who were visited daily.

2. For the 39% figure, let's try something like a PICO, but more like a PECO:

    2a. Who is the Population of interest?

    2b. What group is 'Exposed'?

    2c. What is the Comparison (unexposed) group?

    2d. What is the Outcome? Is this a prevalence or incidence measure?

3. For the 39% figure, what is the corresponding RR? What RR is this?

4. What are the hypothesised mediating (causal pathway) mechanisms?

5. At least from just reading this new article, I feel like the observed association could be partly or wholly explained by confounding factors. Without reading the journal article, can you propose and justify at least one confounder? (Feel free to check the article afterwards...)

6. This was an observational study. To confirm causation, do we need to run an RCT? What is the PICO (Population, Intevention, Comparison, Outcome) for your RCT?

From The Guardian

NEW STUDENTS

There is also a high burden of non-communicable diseases, including diabetes, hypertension and heart disease in India; 11.4% of the population – 101 million people – are living with diabetes, and about 136 million are pre-diabetic..

1. Are these incidence figures or prevalence figures? What is in the numerator and denominator for the 11.4% calculation?

The Lancet study found India’s diabetes prevalence to be higher than previous estimations and showed a higher number of diabetics in urban than rural India.

2. Not specifically related to epidemiology, but I don't care for the word "diabetics" in this sentence. Why not?

Using satellite data and air pollution exposure models, they determined the air pollution in the locality of each participant in that timeframe.

3. If you had unlimited resources, what would be the perfect way to measure participants' exposure to air pollution? Comment on the risk of bias for the way the researchers actually did the exposure measurement. Is this good enough?

They found for every 10μg/m³ increase in annual average PM2.5 level in the two cities, the risk for diabetes increased by 22%.

4. What is the relative risk figure which corresponds to 22%? What is the name of this relative risk?

5. Do you think these findings are generalisable beyond India?

RETURNING STUDENTS

In the BMJ study, the researchers followed a cohort of 12,000 men and women in Delhi and Chennai from 2010 to 2017 and measured their blood sugar levels periodically.

6a. What kind of cohort study was this? 

6b. What is the exclusion criteria for this study?

6c. In the study do you think their outcome was diabetes incidence or diabetes prevalence?

“This study is an eye-opener because now we have found a new cause for diabetes that is pollution.”

7. What is the mediation/causal pathway for PM2.5 to cause diabetes?

8. Is a randomised control trial required to prove PM2.5 causes diabetes?

From the New York Times

A recent research review on exercise and depression found that adults who got the widely recommended 2.5 hours of moderate physical activity per week had a 25 percent lower risk of depression compared to people who didn’t exercise at all. But those who completed just half of the recommended 2.5 weekly hours still had an 18 percent lower risk of depression compared to people who didn’t exercise.

FOR NEW STUDENTS

1. What is the outcome and what are the comparison groups?

2. What are the RRs which correspond to the figures 25% and 18%?

FOR RETURNING STUDENTS

3. What is the mediator (causal pathway) hypothesised?

For example, a 15-year study on over 55,000 Americans ages 18 to 100 found that running just five to 10 minutes per day at a slow pace (under six miles per hour) was associated with “markedly reduced risks” for all causes of death.

4. What is the study design, and why was this design preferable to other options?

5. Aside their outcomes, are there differences between people who run and people who don't? Why do some of these differences matter?

6. To prove causation, do we need to "run" a trial?

From The Guardian.

The results revealed that 730 people were diagnosed with adult ADHD over the study period, 96 (13%) of whom were also diagnosed with dementia. By contrast, there were 7,630 dementia diagnoses (7%) among those who did not receive an adult ADHD diagnosis.

For new students

1. In this study, is dementia measured with prevalence or incidence?

2. What are the eligibility criteria to be in this study?

3. Calculate and name a crude (unadjusted) relative risk.

For returning students

4. What are key advantages of using existing medical records to conduct a study? And the key disadvantages?

After taking into account factors including age, sex, socioeconomic status, smoking and various health conditions...

5. What does it mean to take smoking 'into account'? Was this a justifiable decision?

6. What was the effect modification the investigators found?

7. What mediating (causal pathway) mechanisms did the investigators hypothesise?

However, Prof Chris Hollis, of the University of Nottingham, said there could be a number of factors muddying the waters. “Those adults who seek and receive an ADHD diagnosis are also more likely to be assessed for other cognitive/neuropsychiatric conditions including dementia,”

8. What epidemiologic concept is Prof Hollis implying here? Explain the implication.

9. This study was done with medical records from one country. Does that mean the findings are not generalisable to other countries?

10. Does this hypothesis require an RCT to confirm causation?

From the New York Times

If the link doesn't work, search for "Do Early R.S.V. Vaccine Trials Have a Henrietta Lacks Story?" by Charles Blow.

He poses the question:

Welcome to Epi in the News for the 2023/24 academic year!

This week's article comes from The Guardian.

1. Is the 14% a prevalence or incidence figure?

2. Who is in the numerator and denominator to calculate the 14%?

3. How was the case status 'addiction' defined?

4. To what extent is this problem a public health priority?

And for returning students, here's a bonus article from The Guardian.

A fascinating bit of history from New York Times. 

This is probably the last post from Epi in the News for the 2022/23 academic year. It's been a pleasure sharing these stories with everyone. I hope EITN has helped you develop as an epidemiologist.

Good luck to those who are submitting their MSc projects!

From The Guardian

Researchers compared the medical records of 24,624 people in the US with Parkinson’s, 19,046 people with Alzheimer’s and 23,942 people with cerebrovascular disease.

Those with Parkinson’s were matched with patients in the other groups for age, sex, race and ethnicity, and length of diagnosis to compare the frequency of gastrointestinal conditions in the six years before diagnosis.

1. From the Guardian..

- How would you determine whether the practice of chhaupadi had been eliminated in an area?

2. From the Independent..

- What is the study design? What are the strengths of this design for this hypothesis?

- Break down the PICO.

- What interaction was analysed?

"The epidemiologist crime-stopping duo, Ebonee and Retta, start off office-bound, then upgrade to interrogating sources and slinging threats like hard-boiled detectives."

I welcome your comments about this review (or your own review, if you've read the book) or suggestions of other books of interest to epidemiologists. I can think of plenty of non-fiction books, but nothing comes to mind in fiction...

A rare post about an infectious disease model, in The Guardian.

"In a population with no immunity, a single case of measles will infect between 10 and 20 others.

To maintain herd immunity, the World Health Organization set a target of 95% vaccination uptake."

Explain how these two statements are mathematically related.

A bit too late for those of you taking Paper 2 is this episode of BBC 4's More or Less.

From The Guardian

I was struck by this sentence: “There are misconceptions about using epidemiological evidence for proving causation that should be addressed in a court of law."

What do you think it means?

What sort of (epidemiologic) evidence would you want to have to prove that an exposure was causing birth defects?

From The Guardian

1. What is the study design?

2. What are eligibility criteria?

3. What was the primary outcome definition? And secondary?

4. What were the comparison groups?

5. Calculate and interpret a relative risk figure.

6. Calculate and interpret a NNT (number needed to treat).

7. Do you believe that a causal relationship has been confirmed?

Overlooked No More: Alice Ball, Chemist Who Created a Treatment for Leprosy

First, an article about fairness, equity and diversity in research from Nature.

Second, a podcast from KQED in the San Francisco Bay area. A regular occurance during the many years I've been running Epi in the News are research findings about unequal outcomes for births by ethnicity. This is the first time I've heard about an intervention!

1. What is the intervention?

2. What are the outcome(s)?

3. What are the hypothesised mediating factors?

4. How would you evaluate this intervention? Is an RCT necessary to prove causation?

5, Would findings from Alameda County/California/the US be generalisable to the UK, where there is also a notable inequality for maternal mortality?

From the Guardian

1. What is the exposure? Comment on how the exposure was measured. Can you suggest a more accurate method?

2. What is the outcome? How was this measured?

3. What is the RR reported in the article, and the comparison groups?

4. What are the hypothesised mediators? How would you assess this mediation in a future study?

5. What confounders were accounted for in the analysis? Can you justify consideration of other potential confounders?

6. Is an RCT required to prove causation? If so, how would you design it?

For those who aren't in the know, the BBC More or Less podcast is a terrific source of epidemiology content.

A recent episode had a segment about using purchase data from Boots loyalty cards to predict ovarian cancer diagnoses.

1. What was the study design?

2. What epi measurements would you use to assess the usefulness of the prediction?

3. Could we test this prediction in an RCT? How?

From The Guardian

1. Have you heard about guinea worm before? If not, why not? (Learn more here)

Only 13 cases of guinea worm disease were reported worldwide in 2022, a provisional figure that if confirmed would be the smallest ever documented, the US-based Carter Center has said.

2. Is this a prevalence or incidence figure?

3. You've learned how to design studies to measure prevalence or incidence. But how would you design a study to measure these in a population when the hypothesised value is 0.0?  How do you prove an infection has been eliminated?

The remaining endemic countries are Chad, where six of last year’s human cases occurred; South Sudan, which recorded five; Ethiopia, which saw one; and Angola, Mali and Sudan, which recorded no cases. The Central African Republic, a non-endemic country, reported one case, which is under investigation.

4. If you were the Minister of Health of Chad or Ethiopia, could you justify the national guinea worm programme as a public health priority? Surely the money could be spent better elsewhere?

From The Guardian

1. What is the PICO for the new study?

2. How is the intervention allocated?

4. Comment on the blinding/masking aspect of the control arm.

4. What inputs are required to get the sample size of 280?

5. What is the hypothesised mediating factor / mechanism of action?

6. Does clinical equipoise exist for this study to be proposed?

7. If you're on the ethics board which reviews this study, what questions would you have for the researchers?

This should be it for Epi in the News in 2022. Next year I'll send on articles, though less regularly, and perhaps without questions. I hope you've enjoyed these articles, and have learned to read the newspaper with a bit more a critical eye.

From The Guardian

1. What was the exposure? How was this coded?

2. What was the outcome? Is this an incidence or prevalence figure?

3. What was the hypothesised mediating factor?

4. What is the study design?

A large study that involved 79,952 US-based men found that those who ate the largest amounts of healthy plant-based foods had a 22% lower risk of bowel cancer compared with those who ate the least.

From The Guardian

1. What is the study design?
2. What is the numerator for the maternal mortality rate calculation?
3. What is the denominator for the maternal mortality rate calculation?
4. Is the maternal mortality rate really a rate, in the epidemiologic sense?
5. Is the maternal mortality rate an incidence or prevalence figure?
   

Mothers in the UK are three times more likely to die around the time of pregnancy compared with those in Norway, according to an international analysis of data.

5. Explain how the ‘three times’ figure was calculated. What is the name of this epidemiologic measure?
   

FOR RETURNING STUDENTS

6. What is the benefit/usefulness of presenting unstandardized maternal mortality rates?
7. What is the benefit/usefulness of presenting standardized maternal mortality rates? What variable(s) should be used to standardize?

Only a couple weeks until World AIDS Day, and with that comes increased media attention around sexual health.

One article from the New York Times. (If the link doesn't work, search for "Bringing Sexy Back — To Fight H.I.V." for 15 Nov 2022)

That is what the World Health Organization and a small nongovernmental organization called the Pleasure Project found when they reviewed the results of safer-sex trials and experiments over the past 15 years.

And another from The Guardian, though I can't come up with many questions for this article, as it seems to be a miscellaneous collection of statistics.

It highlights that the number of STIs recorded among over-65s increased from 2,280 in 2017 to 2,748 in 2019 – a 20% rise.

2) The implication of this sentence is that the observed increase in reported STI diagnoses reflects an actual increase in STI cases. What are some plausible, alternative explanations for the observed increase?

From the New York Times. Search for "Scientists Say Concussions Can Cause a Brain Disease. These Doctors Disagree".

This is an interesting article about the association between concussions and brain disease, and whether there is enough evidence to declare that the association is actually a cause.

I welcome your comments.

Hello,

I was thoroughly confused by this website  https://www.gov.uk/guidance/case-control-study-comparative-studies which describes a paper (details below) where they call cases post-operative bariatric surgery patients who have telemedicine technology and controls post-operative bariatric surgery patiints who do not and then looks at how well they attend for follow up.

I had thought that case/control status had to be an outcome, and that aim of a case control study was to examine the exposure. In this study it seems to me that the case/control status is an exposure and they are examining an outcome. I would describe this as an observational cohort study. What do others think?

Thank you

Judith 

From The Guardian

The UK ranks among the worst in Europe with the most overweight and obese adults, according to the World Health Organization. On obesity rates alone, the UK is third after Turkey and Malta.

1. What epidemiologic statistic do you think the WHO used to rank countries?

There is a seven times greater risk of type 2 diabetes in obese people compared with those of healthy weight, and a threefold increase in risk for those just overweight.

2. What is the value of the relative risk (RR) figures estimated here? Who are in the denominators and the numerators? What is the specific name for this RR?

3. (For returning students) What sort of epidemiologic study design was most likely used to estimate these RRs? Why was this design more likely than the others?

The number of people under 40 in the UK diagnosed with type 2 diabetes has jumped 23% from about 120,000 in 2016/17 to 148,000 in 2020/21, according to Diabetes UK.

From the Guardian

FOR ALL STUDENTS

1. What is the outcome? Is this a prevalence or incidence figure?

2. What is the study design?

3. Calculate and interpret RRs for feeling safe in the North and the Midlands, using the South as the reference group.

4a. Calculate and interpret an RR for feeling safe for white girls, using girls of colour as the reference group.

4b. Calculate and interpret an RR for feeling safe for girls of colour, using white girls as the reference group.

FOR RETURNING STUDENTS

5. Comment on how sampling bias might have affected the 19% figure.

6. Comment on how measurement bias might have affected the 19% figure.

7. Comment on whether the RR figures are true and meaningful?

Bonus entry for this week! From the Guardian.

FOR ALL STUDENTS

“We estimated that 1.64% of women who never used hair straighteners would go on to develop uterine cancer by the age of 70, but for frequent users, that risk goes up to 4.05%,” the study leader, Alexandra White of the US National Institute of Environmental Health Safety (NIEHS), said in a statement.

1. Are these incidence or prevalence figures? Who is in the denominator and numerator for these calculations?

2. Calculate a relative risk figure.

3. What are the exposure categories implied in the article? How do you think the researchers measured exposure?

...the odds of developing uterine cancer were more than two and a half times higher for women who had used straightening products more than four times in the previous year.

4. What specific type of relative risk did the researchers calculate?

5. Assuming the researchers correctly estimated the causal effect here, what proportion of uterine cancer experienced among people who use straightening products frequently was attributable to the hair straightening? What proportion of these people would have experienced uterine cancer anyways?

FOR RETURNING STUDENTS

5. What mediator (causal pathway) is mentioned in the article?

6. The researchers assessed effect modification. What does this mean in terms of the RRs they estimated?

7. Do you think the RR results were affected by misclassification of exposure? Would this be differential or non-differential?

8. How could the researchers argue that they found a causal relationship?

A detailed and fascinating read from the New York Times.

If the link doesn't work, google search "‘Something Is Wrong Here.’ More Than 100 Local Residents Had Brain Tumors, but What Was the Cause?"

Search "Fentanyl From the Government? A Vancouver Experiment Aims to Stop Overdoses" for the article from the New York Times.

1. How would you evaluate this programme? What would you use as the 'exposure' and primary 'outcome' measures?

1b. What about secondary outcomes and adverse events?

2. What do you think about the mechanism of action / mediators?

3. Could you design a trial to evaluate the intervention?

It's become a truism that epidemiologic research about the effects of coffee or chocolate, will be reported in the mass media.

Here's something from The Guardian.

1. What is the study design?

2. What are the exposure / comparison groups? How were these data collected? Is this subject to measurement bias?

3. What is the outcome? How was this collected? Is this subject to measurement bias?

4. Explain why 'ethnicity' was part of the analysis? Would it be a problem if the ethnicity data was missing?

5. Explain what is the numerator and denominator for the calculation leading to the statement "29% lower risk of death".

6. Is an RCT necessary to demonstrate causation? Is this feasible?

From the New York Times

From The Guardian

The study, published in the journal Plos One, analysed surveys completed by 2,536 dog owners about a single animal. Just over half ate conventional meat-based diets, a third were fed raw meat and 13% had vegan diets.

1. Are the figures 'half' 'third' and '13%' incidence or prevalence figures? What numbers are in the numerators and denominators for these?

Among the findings were that 17% of dogs on conventional diets had four or more visits to the vet over the course of a year, compared with 9% for those on vegan diets and 8% for those on raw meat diets.

2. What is the outcome definition? Why was it defined in this way?

3. What sort of study design is implied with these outcomes? What are the strengths of this design, for this particular research question? And what are the limitations of this design?

4. There is a mediator hypothesised here, what is it?

5. There is a confounder suggested here. What it is, does it qualify as a confounder? Can you think of other potential confounders?

Further research is needed to confirm the findings. “The key limitation of our study is that we didn’t have a population of animals locked up in a research facility and fed one specific diet without any alteration,” Knight said. “We studied what real dogs in normal homes ate and their health outcomes. It gives us a good indication as to what the outcomes are for dogs in the real world.”

6. What study design is implied here? Do you think this is necessary (or feasible)?

Most of the respondents to the survey were in the UK and other European countries and more than 90% were women, but Knight said this was unlikely to have caused a systematic bias.

7. Do you agree with Knight?

From The Washington Post

1. Is investigating this topic a public health priority? How would you get evidence to make this decision?

2. What is the exposure (or comparison groups)?

3. What outcomes are listed here?

4. For one outcome you listed above, what do you think are potential mediators for the exposure-outcome relationship?

5. What would be your first choice for an epi study design would you use to investigate this relationship further? Why?

From Slate

Only one question this week: What is your reaction to the RCT that Brett Worly proposes?

This issue also reminds me of a hypothesis that arose a few years ago, that certain forms of hormonal birth control were associated with increased susceptability to HIV. Again, the means to come up with a way to get a definitive answer - balancing both epidemiologic and ethical considerations - is extremely challenging. But yet extremely important to resolve, given the wide use of hormonal birth control.

I welcome your comments! -s

Not news, but something from a newsletter. The Epidemiology Monitor, which is for epidemiologists (and tends to focus on North America).

From Nature

Dracunculiasis, to me, is the most interesting infectious disease out there. Caused by the guinea worm, it is a target of an eradiction effort, and may become the second disease ever eradicated among humans.

The International Task Force for Disease Eradication currently has eight diseases identified as potentially eradicable. In addition to Guinea worm, these are poliomyelitis, mumps, rubella, lymphatic filariasis, cysticercosis, measles and yaws.

1. What are the characteristics of diseases which make them potentially eradicable?

2. Check out the graph 'On the Way Out'. Explain why the Y axis is presented this way.

3. How would you determine that the incidence of a disease is 0?

4. How would you track the incidence or prevalence of infection among animal hosts?

From The Guardian

1. What are the exposure categories? What are the outcome categories? What is the null hypothesis?

3. What are some reasons why a study can end with a finding of "no statistically significant increased risk"?

4. How would you find evidence to demonstrate a causal relationship, if one were to exist?

Cataract surgery may reduce your dementia risk - from the New York Times

I'll throw this over to you. Can you figure out what questions I would ask from this article?

From the New York Times

1. What was the incidence risk of MS? Is this common?

2. What is the exposure? Is it common?

3. What is the study design?

4. Wait a second. Shouldn't we do case-control study?

5. Explain how temporality might be an issue.

6. Do we need to do a randomized trial to establish causation?

From The Guardian

From The Guardian

1. What study design is most suitable for investigating a potentially new disease of unknown cause?

2. What is the outcome definition? What is the implication of having a ambiguous definition on the search for causes?

3. Would an RCT be required to confirm the cause?

And a bonus article from The New York Times - notable for some excellent examples of data visualisation, and for hitting on most of the main points raised in the 'screening' session.

1. How can a test be 99.9% "accurate" and yet have very poor positive predictive value?

2. Under what conditions should we routinely screen for extremely rare conditions?

Hello everyone - thank you for a wonderful start to the academic term. This is the last Epi in the News post for Term 1. I'll continue to post next year, though less frequently.

Enjoy your holiday break! -sujit

From The Guardian

1. What is the outcome? Why is this outcome important for public health?

2. What are the exposures of interest?

3. What do you think about the way the exposures are defined? What's the potential for bias?

4. What is an advantage of collecting data via app, rather than a questionnaire?

5. What is the study design?

But when the researchers took age, ethnicity, education, and occupation into account,

6. What is the epi concept? Explain why occupation is included.

7. What are the hypothesised mediating factors between exposure to natural spaces and loneliness? Do you agree?

the benefits of nature contact and feelings of social inclusion on loneliness remained strongly statistically significant.

7. How would you explain this sentence to a non-epidemiologist?

8. Should we do a randomised trial to confirm these findings?

From The Guardian

Women in prison are five times more likely to have a stillbirth and twice as likely to give birth to a premature baby that needs special care, new data collected by the Observer shows.

1. What are the exposure (comparison) groups?

2. What are the ratio measures (in terms of the numeric figures) which are referred to here?

3. Are these incidence or prevalence figures?

A higher rate of drug and alcohol problems within prisons than in the general public could be one of many contributing factors to the poorer birth outcomes.

4. What's the epidemiologic concept refered to here? How would you deal with this situation in analysis?

Prisons are extremely stressful and traumatising environments that can affect a mother’s pregnancy, said Paradine.

5. What's the epidemiologic concept refered to here?

6. Can we do a trial to establish causation?

Hello everyone - it's been a slow news week, but I spotted this in The Guardian

1. What is the study design?

2. What are the exposure / comparison groups? Who is used as the reference group?

3. What are the outcomes?

4. What mediators have been proposed?

5. The article makes no reference to confounders. Can you suggest one important confounder, and explain why?

6. Could this research question be investigated with a case-control design?

7. Do we need a trial to demonstrate the association is causal? Can you think of one advantage and one disadvantage of doing a trial?

Finally - here's an obituary for Zena Stein in the New York Times.

From The Guardian

1. What is the study design?

2. What is the exposure? How could this measure be biased?

3. What is the outcome? How could this measure be biased?

4. Is suicide an example of a prevalence measure or an incidence measure?

5. Is "increased discomfort" a confounder?

6. What effect modifiers did the researchers find?

From The Guardian

1. What is the epidemiologic study design?

2. What would be the key eligibility criteria to be part of the analysis?

More specifically, those who fell asleep at midnight or later had a 25% higher risk of going on to develop cardiovascular disease, while those who fell asleep before 10pm had a 24% increased risk.

From The Guardian

1. What is the epidemiologic study design?

Asked how they felt over the previous two weeks, 37% showed “moderate to severe symptoms of depression” and 39% showed signs of “likely having some form of anxiety”.

2. Are these figure prevalence, prevalence odds, incidence risk, incidence rate?

The ONS cautioned that the statistics were “experimental”, based on a relatively small sample of about 2,000 first- and foundation-year students in English universities who were invited to take part via email.

3. What are the potential issues with sampling bias, and what are the potential implications of this bias on the study findings?

The survey also found that their satisfaction with life was significantly lower than that of the general adult population, at 6.6 out of 10 compared with 7.1, though similar to the general student population at 6.5.

4. How could the 'satisfaction' results be affected by confounding? What are some potential ways to address the issue?

5. What does "significantly lower" mean, statistically?

From the New York Times

NB: Q8-11 are for students who have already completed Basic/Extended/EPM101.

1. What were the outcomes of interest? Were these prevalence or incidence figures?

2. What was the exposure, and how was it measured?

3. What is the study design? (Ok, actually, designs)

Compared with those in the lowest quarter for cumulative exposure, those in the highest had a 57 percent increased risk.

4. How was the exposure classified?

5. How did the researchers set up a calculation to get to "57 percent increased risk"?

...their study had too few cases of stroke to achieve statistical significance.

6. What can you assume about the 95% CI for the association between lifelong LDL and stroke?

7. Does this mean there is no association between lifelong LDL and stroke?

8. What epidemiologic research design is (usually) more suitable for investigating risk factors for rare outcomes such as stroke?

The study controlled for race and ethnicity, sex, year of birth, body mass index, smoking, high-density lipoprotein (HDL, or “good” cholesterol), blood pressure, Type 2 diabetes and the use of lipid-lowering and blood pressure medicines.

9. Why did the researchers control for smoking?

10. Why did the researchers control for lipid-lowering medicine? (I actually don't know - this seems strange to me!)

11. Can we conclude that the researchers found some causal relationships here? Or do we need an RCT?

From The Guardian.

NB: You can answer all of these questions using the only text of the newpaper article.

1. What was the study design?

1b. Why can't you investigate this hypothesis with a cross-sectional design?

2. What were the exposure categories?

2b. Why did the researchers not use a placebo control?

3. What was the outcome of interest?

4. Was the outcome measured as a prevalence, incidence risk, or incidence rate? Why was this a sensible choice?

5. There's evidence of effect modification implied in the article. Can you find it? (Most of you wont be able to answer this yet)

6. What are some of the mediating factors for this exposure-disease relationship? (This question is beyond the scope of our modules, but I'll ask anyways)

From The Guardian

The headline is "Nearly 25% of world population exposed to deadly city heat"

1. Is this a prevalence or incidence figure? What is in the numerator and the denominator?

...Bangladesh’s capital experienced an increase of 575 million person-days of extreme heat.

2. How would you explain this sentence to a non-epidemiologist? Why is a figure like this useful, compared to a figure like 25% from the headline?

“It increases morbidity and mortality. It impacts people’s ability to work, and results in lower economic output. It exacerbates pre-existing health conditions.”

3. The author is making a causal statement about heat. What sort of RR value would you expect for the association between heat and mortality? How would this RR be calculated (what goes in the numerator, what goes in the denominator)?

4. How would you estimate the proportion of deaths in a city that is attributable to extreme heat?

From The Guardian

In the 2019 school year, 23.4% of five-year-olds in England and 26.5% of four- to five-year-olds in Scotland had experienced damage to their teeth.

1. Is this an incidence figure or a prevalence figure? What's in the numerator and denominator for this calculation?

Indigenous children, many of whom live in communities without fluoride, had a 70% rate of tooth decay. The rate was 55% among all Queensland children aged between five and 15.

2. Can you suggest the appropriate epidemiology word to replace "rate" here?

Chris Whitty, the chief medical officer for England, and his counterparts in Wales, Scotland and Northern Ireland cited estimates by Public Health England that adding more fluoride to water supplies would reduce cavities by 17% among the richest children and 28% among the poorest.

3. How did Chris Whitty calculate these figures?

4. (For those who have completed their first epi module) What's the fancy epidemiologic concept that should come to mind when you see the figures 17% and 28%?

From Science

No questions, but I welcome your comments.

A long article in the New York Times about a cluster of pediatric HIV cases in rural Pakistan, with no evidence of vertical (mother-to-child) transmission.

In many ways, the public-health system in Ratodero is like public-health systems everywhere: Its workers are understaffed, underpaid, disillusioned. The work is tedious, and the reward for success can be invisible. After all, the public doesn’t realize when disease is prevented; it only knows when it’s not. Governments need to keep an accurate count of cases, track where and how a virus is circulating and coordinate a response to choke its spread — or at least slow it down. Even the most heroic efforts by individual doctors and nurses aren’t substitutes for government leadership and public-health action. When they’re inadequate, preventable outbreaks erupt, the difficult-to-control turns impossible. Diseases unfurl. People die.

From The Guardian

1. How is the exclusion rate calculated?

2. What are the comparison groups? Which group is the baseline?

3. How would you set up an equation to be able to make the finding "up to six times higher"

4. The relative rate of exclusion Black Caribbean students compared to white students varies by local authority. What is the fancy epidemiologic term for this phenomenon?

"Figures show that in Cambridgeshire the fixed-term exclusion rate for black Caribbean pupils was more than six times higher than the rate for white British students, while in the London boroughs of Brent, Harrow and Haringey, the rate was more than five times higher. Though Cambridgeshire has a relatively small number of Caribbean students, which partially explains the disparity, Brent, Harrow and Haringey have significant Caribbean populations."

5. Does the above sentence make sense? I'm struggling!

6. How would you design a study to determine whether exclusion had long-term impact on students' social and economic well-being?

From The Guardian

1. What is the null hypothesis for this research?

2. What sort of epidemiologic study design would be appropriate?

3. What is the measurement scale for the exposure? Outcome?

4. What (classic) statistical test would you use to test the hypothesis?

From the Guardian

1. What is the study design?

2. Can you come up with a PICO breakdown? To what extent is this study generalisable to other populations?

3. What is notable about the figures 28% and 32%? (Vague question - I know!)

4. What does "statistically significant improvements" in emotional health mean?

5. Did this study demonstrate a causal effect?

6. One thing I like about this article is that there is a mechanism of action (otherwise known as the causal pathway) described for one of the outcomes. What is it? Why is it important to consider the mechanism?

From The Guardian

1. What is the outcome of interest in this pilot study?

2. How would you determine whether self-testing is as accurate as clinical testing? What statistics / metrics would you use?

3. What are potential disadvantages of routine self-testing? How can these be mitigated?

From The Guardian

1. How would you determine whether botched Brazilian butt lifts are a public health problem? Is there enough evidence to support a change in surgical regulation?

2. How would you identify which groups of people are more likely to get the surgery?

3. How would you identify the risk factors for major complications?

"..free or low-cost cosmetic procedures are still available in the public health system."

4. How would you convince a policy maker to cover the Brazilian butt lift under the national insurance system?

I'm stepping away from 'in the news' just for today because I thought it was worth sharing these two resources:

An article from David Verga of PATH "How we talk about public health and why it matters"

I'll admit having used many of these phrases, and am sympathetic to his arguments.

A opinion piece from Lioba Hirsch of LSHTM "Is it possible to decolonise global health institutions?"

This one is more challenging, and certainly more provocative.

I welcome your thoughts! Are there suggestions here you find worthwhile? Or that are infeasible or unreasonable? How can our practice of public health (and our teaching) improve?

From The Guardian

The survey questioned more than 110,000 people around the globe, including 5,283 in the UK, in a three-month period from November 2019 to February 2020, before the coronavirus pandemic.

1. What is the study design?

More than 5% of people under 25 in the UK reported having sought hospital treatment after getting drunk, compared with a global average of 2%.

2. What is in the numerator and denominator of these figures?

Respondents were asked to say how many times they had got so drunk that “your physical and mental faculties are impaired to the point where your balance/speech was affected, you were unable to focus clearly on things, and that your conversation and behaviours were very obviously different to people who know you”.

3. What are the strengths and limitations of the definition of "drunk" ?

I'm just going to leave this here: https://www.globaldrugsurvey.com/about-us/methods-and-limitations/

Something nice and short from the New York Times. I will restrain myself from going on protracted rant about nutritional epidemiology.

1. What is the study design?

2. What would be the key inclusion criteria to be in the study? How plausible is it that the researchers got this aspect of recruitment right?

3. How many exposure comparison groups were there? Comment on how these groups were defined.

Those in the highest one-third in consumption of both vitamins together had a 38 percent reduced risk.

4. What is the RR?

5. Why wasn't the RR for the middle one-third reported? (I'm asking you to speculate!)

6. Are you convinced? Or do we need a trial?

7. Calculate an overall incidence figure, and then report in a way that your non-epidemiologist friends would understand.

From the New York Times (business section!)

"One in five adults was experiencing depression, bipolar disorder, schizophrenia, post-traumatic stress or some other malady, according to the National Institute of Mental Health. The rates were significantly higher for adolescents (about 50 percent) and young adults (about 30 percent)."

1. What epi study design is used to make this measure?

"For instance, exposure to nature has been shown to lower cortisol levels, a measure of stress."

2. What study design is used to make this conclusion? How would you measure the exposure?

"If patients are less stressed, they may make faster and more lasting progress during treatment, experts say."

3. How do experts know this?

"Paths will be lined with cedars and pines, rosemary and lavender — plants whose scents activate “natural killer” cells that can strengthen immunity, said Richard Dallam, a managing partner at NBBJ and a leader of the firm’s health care practice."

4. Can someone fact-check this??

5. How would you design a study to test whether building design affects recovery from severe mental illness?

From the New York Times

The practice of epidemiology is often dominated by the "determinants" aspect, with less priority/prestige on the "distribution" side. I appreciate this article because of how a very simple descriptive (distribution) statement sparks a lengthy discussion of potential determinants.

"In data released in 2017, the rate for white Americans was around 19 per 100,000, and it was about 7.1 for both Hispanics and Asian-Americans/Pacific Islanders, and 6.6 for Black Americans, according to the Centers for Disease Control and Prevention."

"Suicide rates are highest among Native American and Alaska Native populations: 21.8 per 100,000 people"

1. Let's say you are in charge of a funding agency. What hypothesis mentioned here is the one for which you would prioritise funding for a study? Why?

1b. What are the modifiable risk factors mentioned in the article?

2. At the end of this article (and indeed any article concerning suicide in reputable, mainstream print media), editors have appended a statement for self-referral. I've included the statement used for articles published in The Guardian, below. You won't see statements like this with articles about other health conditions. Why?

In the UK and Ireland, Samaritans can be contacted on 116 123 or email jo@samaritans.org or jo@samaritans.ie. In the US, the National Suicide Prevention Lifeline is 1-800-273-8255. In Australia, the crisis support service Lifeline is 13 11 14. Other international helplines can be found at www.befrienders.org.

From The Guardian

1. In response to this report, can you come up with (at least) one specific hypothesis to investigate?

2. What is the exposure? The outcome? How would you measure these?

3. What is the theoretical mechanism of action, and how can you test this?

“Harmful” gender stereotyping has helped fuel the UK mental health crisis afflicting the younger generation, an influential report has warned, adding that it is at the root of problems with body image and eating disorders, record male suicide rates as well as violence against women and girls.

4. The authors of the report are making a causal argument about gender stereotyping. How do you think they arrived arrived at this conclusion?

From The Guardian and BMJ Christmas issue.

"...they discovered that owning a dog with diabetes was associated with a 38% increased risk of having type 2 diabetes compared with owning a healthy hound."

1. What is the RR figure for 38%? Explain what is in the numerator and the denominator to calculate the RR.

2. The word "having" is incorrect. What is the correct word? Why does this matter?

3. After looking at the BMJ article, I see that the Guardian journalist should not have reported the 38% figure. Why?

4. What are the ethical aspects involved with data collection for this study?

The incidence of diabetes in the pets was 1.3 cases per 1000 dog years at risk and 2.2 cases per 1000 cat years at risk.

5. How would you explain the above figures to a non-epidemiologist pet owner?

6. What do you think are the important confounders to consider, and why?

7. Related to #6, is a causal relationship possible?

8.  Might this association be extended to hypertension? Educational video here

From the Times of India

1. What study design should we use to identify the cause(s)? Can you explain why other epidemiologic designs aren't suitable?

2. What are the first few decisions epidemiologists should make as they design this study?

From Vox

No questions this time, but I'm keen to hear your opinions about this approach, and if/how it should apply in health research.

I'm a subscriber to The Epidemiology Monitor, which is a newsletter of interest to epidemiologists. It has a USA focus, though not exclusively.

In their latest newsletter they are offering free epidemiology books to readers in the USA

https://issuu.com/theepidemiologymonitor/docs/final-november-2020-the-epidemiology-monitor

I'm stretching the definition of "in the news" here, with a brief article from MMWR in 1981. Do read through the Editor's note as well.

1. As more patients are identified as part of this outbreak, what study design would you use to identify the cause?

2. An early hypothesis was that sexual orientation was the cause. Knowing what we know now, what confounders should have been considered?

From the New York Times.

For your reading pleasure.

From Science 

Not an epidemiology study per se, but there is an exposure, an outcome, and an effect. You should also be able to read through and the identify key epidemiology concepts. For example, I spotted an example of effect modification...

No questions this time. I'll ask you to put your peer reviewer hat on and see if you can spot the limitations.

I'll ask everyone who posts to identify ONE limitation, and explain the implication on the finding.

And then, if you are feeling energetic, to explain how you would design a study to measure the effect of a mentor's gender.

UPDATE:
https://retractionwatch.com/2020/12/21/nature-communications-retracts-much-criticized-paper-on-mentorship/

From The Guardian

From the first sentence of the second paragraph: "The findings showed that the climate crisis is taking a toll on the mental health of young people."

Setting aside the question of whether climate change is a public health problem*, I want to interrogate whether this study shows that eco-anxiety is a public health problem.

1. What are the limitations of this study?

* Yes, it is.

A double-bill from The Guardian for you:

First, an article about video games and mental health.

1. What is the study design?

2. Comment on the exposure definition

3. Comment on the outcome definition

4. What was the finding? How can this finding be used to inform policy?

Second, an article about the nocebo effect, with a quote from LSHTM's own Prof Liam Smeeth.

1. What is the study design?

2. Are the researchers able to demonstrate a causal relationship?

3. Comment on the quality of reporting for the first versus second article.

A long post-mortem from the New York Times about election polling in the United States.

1. What epi study design is an election poll?

I'll encourage you to read through and pick out where they are talking about selection bias and where they are talking about measurement bias. There's no shortage of epi-related content here!

2. How would you improve polling for the next election?

From the New York Times

"The randomized, double-blind clinical trial was conducted by the H.I.V. Prevention Trials Network, an international collaborative funded by the National Institutes of Health. The trial compared the injected drug, called cabotegravir, with Truvada in 3,223 participants in 20 sites across seven countries in sub-Saharan Africa."

1. What is the primary advantage of an RCT over observational epi study designs?

2. What does "double-blind" mean? (ok, this is a trick question - read the CONSORT checklist guidelines to learn more)

3. Why was Truvada allocated in the control arm, and not a placebo?

"After an interim analysis showed that the long-acting injection was 89 percent more effective than Truvada, an independent data safety monitoring board recommended that the trial be stopped early."

4. What is the RR for this study?

5. Why was the trial stopped early?

"A previous trial tested the drug in nearly 4,600 cisgender men and transgender women who have sex with men and found it to be 66 percent more effective than Truvada in that population."

6. Given evidence of effectivness from the previous trial, why was a new trial conducted with cisgender women?

7. Assuming an affordable price, would you recommend cabotegravir for routine use among high-risk individuals in your country?

8. Will the efficacy of cabotegravir be the same in another population of cisgender women? What about the effectiveness?

From the New York Times

I'll start with a quote from the end of the article: “They have a Solomonic decision to make, with one study that demonstrated very promising effects and the other study that didn’t demonstrate an effect,” he said. “I think it’s a challenging decision, because everybody wants to do what’s best for patients and families.”

1. What sort of evidence would you want to have, to determine that this drug can cause a delay in cognitive decline?

2. What are the pros and cons of approving this drug?

From The Guardian

An article with perfect timing for the Basic Epidemiology students, as this week they consider questions of causation.

There’s an awful lot we don’t know about the disease, as an editorial in The Lancet last year said: “Contrary to common perception, the clinical syndrome of CTE has not yet been fully defined. Its prevalence is unknown, and the neuropathological diagnostic criteria are no more than preliminary.”

1. Why is it a problem to not have a defined clinical defition of CTE? Why is it a problem to not know the prevalence?

CISG produced the consensus by pulling together all the available research on CTE published in the previous 10 years. In 2016, they found 3,819 relevant studies. But their criteria for inclusion in the consensus were so strict that only 47 of those studies were accepted.

2.  What are the pros and cons of have strict inclusion criteria for a systematic review / meta-analysis?

A member of CISG said they prioritise longitudinal cohort studies, which study the effects of head trauma in a group of athletes over a length of 10 years or more.

3. What are the epidemiologic study designs that have been de-prioritised by CISG? Would you make the same decision?

4. What evidence would you need to determine that a hypothesised exposure is, in fact, a cause of disease? Is a double-blind, randomized, placebo-control trial required?

From The Guardian

1. Comment on the disease outcome classification. What are the advantages and the disadvantages to measuring mental health status in this manner?

2. From a methodological perspective, what are some potential reasons for the prevalence of mental disorder to increase over time?

3. What findings do you think are examples are descriptive epidemiology? And which are examples of analytic epidemiology?

From the New York Times

This is also an opportunity to share a book review by the late, great Warren Winkelstein, Jr.

"It became apparent during the 1920s that many dial painters were dying prematurely and were suffering from a variety of acute and chronic diseases. Particularly frightening was the frequency of disfiguring cancers and osteomyelitis of the upper and lower jaw."

From The New York Times

1. What study design is this?

2. How was the exposure defined? Comment the limitations of this measurement.

3. Calculate and interpret this: 362/4231

4.  How was the outcome defined?

"After controlling for age, smoking during pregnancy, body mass index, a history of asthma and other factors.."

5. Why did the researchers control for these? Would you have controlled for different variables?

"...they found that maternal depression and anxiety during pregnancy was significantly associated with both diagnoses of asthma and poorer lung function in their children. There was no association between childhood asthma and parents’ psychological distress in the years after pregnancy, and no association with paternal psychological stress at any time."

6. Why should this kind of language make an epidemiologist cringe?

7. Can you design an RCT to test this hypothesis?

In this forum, I will post news articles of interest to epidemiologists and epidemiology students.

With the article, I'll pose a few questions for your consideration. Sometime I'll pose a question knowing that there isn't a 'right' answer. Sometimes I'll ask questions so that you can pull together your learnings from across different modules, sort of like what you'll have to do in Paper 2.

Epi is the News is purely for your enrichment and entertainment purposes. In most cases I'm not necessarily endorsing the study in question, and more likely will be asking you to critique it.

Enjoy! Or feel free to unsubscribe.

Sujit Rathod, MSc Epidemiology Content Director

From the New York Times.

One one question today: How would you evaluate an intervention to improve vaccine confidence?