cultivating & crashing

an organic collection of notes, observations, and thoughts

Tag: epidemiology

Draft emails never sent

from early 2015. An interested one, and then a haughty one.

Hi Professor J,

I’ve been looking at this and wondering how it compares to WinBUGS in terms of the Gibbs sampling component.

From: Stata [service@stata.com]
Sent: Wednesday, April 08, 2015 9:40 AM
Subject: (SCL > 6): Stata 14 is here. Ships today. Downloads today.

Stata release 14: Available now

We’ve got Bayesian analysis, IRT, Unicode, and so much more packed into our latest release. See what Stata 14 has to offer.

Bayesian analysis

Bayesian analysis is a statistical analysis that answers research questions about unknown parameters using probability statements. For example, what is the probability that the average male height is between 70 and 80 inches or that the average female height is between 60 and 70 inches? Or, what is the probability that …

Hi Prof. H,

Just one thing that I kept thinking about since the Q&A session: BIC analysis provides several pieces of information, one of them being a model average that is the optimized model for prediction. Another totally separate set of information it provides is an evaluation of posterior probabilities for each variable to be included in the model (ie, model selection). I think we said something that might have lead you to believe they’re the same thing, but that’s not the case.

Advertisements

uh oh

idea for phd: learn how to apply rigorous epidemiology methods to small samples and imperfect data in mental health research

CSEB

Last week I attended the Canadian Society for Epidemiology and Biostatistics Student Conference, the first conference I traveled to present at. Sandro Galea gave a great keynote speech.

  • population health vs. personalized health
  • mismatch of funds spent and health of populations
  • article in Fortune, “Can genes really predict your health?”
    • our health has deteriorated, but our genes haven’t changed, so why all the fuss (and funding) about them?
  • population estimates do NOT allow us to predict individual event inference (a point about personalized medicine)
  • no lone ranger, no silver bullet. obesity is associated with a HUGE causal web, not one isolated factor
  • book written with Katherine Keyes

slides I was interested in

  • health determinants vs. health expenditures
  • social causes of death vs. others (poverty causes same no. as injury)
  • what percentage of your intelligence depends on genes?
    • it depends on your environment
  • need with grapple with health equity vs. efficiency trade-off
    • people of high SES benefit most from changes
  • embrace intellectual and moral challenges of our time: how to change the socio-economic context that influences health?

metaphors

  • goalie is medicine. for every one goalie, ten players that are not goalies are needed to move the ball up the field. OTHERWISE WE LOSE.
  • goldfish in a bowl: can exercise, have safe sex, and eat not too much food, but if the water isn’t changed, it will still die.

I liked this talk because it confirmed everything I believe about the social nature of how society functions. our health is mostly determined by our social environment, and it’s those social factors that are at the crux of making sure populations are healthy. poverty stunts, sickens, and kills people. we must create societies that take care of the least privileged, the most vulnerable. otherwise, rich people will enjoy the advances of health while their neighbours rot in misery.


We had a session in which we spoke to a group of people who work in different epidemiology/public health positions.

National Collaboration Centre (NCC)

  • different centres in different provinces
  • internships
  • funded by Public Health Agency of Canada

data management and data cleaning are 90% of the work that must be done in data analysis. get experience working with horrible data sets.

Public Health Ontario epidemiologists – mailing list, great resources

make yourself stand out in an interview, educate yourself well about the place/department. also, want to be flexible, be a generalist with enough knowledge about a lot of things, not in-depth for one topic (advice is diametrically opposed to that for someone who wants to continue in academia).

working in government

  • BUREAUCRACY. the right thing to do takes 10 years, but it happens! sometimes even in just 2 years.
  • important impact on populations

Kue Young

  • FYI: WHO makes up data where none exists, read the fine print and footnotes of everything you use
  • Global Health Observatory data is cool, check it out
  • beware the reification of stats due to pretty charts and maps (ie they become true because they are visualized)
  • data don’t exist in a social vacuum

notes on epi after school

CRO contract research officer
SAS + R for govt or industry

Fed
Public Health Agency of Canada
Health Canada
Stats Can – health data analysis
Nat’l Defense (!)

Prov
Ministry of Health
Ministry of Social Services
Public Health
Specific agencies

Regional
Health authorities
Public health units
Hospitals (2-3 analysts)

Federal Student Work Experience Program
Field Epidemiology Program (infectious disease, very competitive)

work ethic
responsible
working on team
good writing
problem solving

CIHI, EBRI, OHRI, research institutes

data analyst, researcher, think tank

consulting/consultancies (eg IMS Health)
many in mental health (pharma, tech)
– check for internships
– refs from employers, not so much academia

epimonitor.net/JobBank.htm
job notifications lists on CIHI, OPHE bulletin, CSEB membership list
jobs.gc.ca

Edit (29/01/2016):
http://www.ices.on.ca/
http://www.awb-usf.org/
https://www.facebook.com/socialmedicinenetworkqc/

The epidemiology of drug overdose

From the BBC.

Would be interesting to model the data. I am certain that socioeconomic status would explain some of the variation, but wonder if it would be more so than population density, for instance.

Anything in the name of science

“Medical scientists are nice people, but you should not let them treat you.”

August Bier (1861—1949)

חשוב

Today is an important day because it’s the day I first came across my dream job: field epidemiologist. And it’s the day that I found out that I can work for MSF without becoming a doctor.

I’m happy; I can go to sleep now. Or maybe start learning Arabic.

Notes from Bad Science

Make kedgeree

Read about this intervention http://www.bmj.com/content/334/7595/678.long

Visit badscience.net

https://holfordwatch.wordpress.com/

http://www.healthwatch.co.uk/

http://retractionwatch.com/

http://www.dcscience.net/

kludge
noun, Computer Slang.
1.
a software or hardware configuration that, while inelegant, inefficient, clumsy, or patched together, succeeds in solving a specific problem or performing a particular task.

http://www.ncbi.nlm.nih.gov/pubmed/10626367

Nullis in verba — On the word of no one
Royal Society’s motto

How old is epidemiology?

Seminar by Alfredo Morabia

1. 17th century move toward centralized states, organization of Europe after 30 year war

2. Crisis of Western ideas
Francis Bacon
– biases
– tabulating data
– determinants of longevity including heredity, height and weight, age, diet, behaviour, exercise, housing, medical treatment

Descartes
– only trust “evidence”
– study one determinant at a time
– ignore interactions

Jan Baptista Van Helmont
– “Dawn or the New Rise of Medicine”
– iatrochemistry
– anti Galenist medicine: believed in diseases, separate from the individual, with external causes
– did not believe in blood letting (Van Helmont contest) “how many funerals each of us will have funerals” idea of fraction that will get better will be greater than your fraction that will get better was opposed to all mine will be cured and none of yours will

this is Renaissance thinking; will be implemented in the Enlightenment

3. Data!
Bills of Mortality (death certificates) in England

Bubonic plague – disruption of society and absolutist states
idea: plague has astrological cause or is transmitted by miasms

person would diagnose and count plague deaths, then publish them weekly
idea was to have a retreat outside of London when there was an outbreak
solution to the crisis every time there was an outbreak, based on weekly death counts, then add other causes of death, then births

4. Emergence of political arithmetic and capitalism
birth of epi is related to birth of capitalism

only merchants could deal with numbers, or intellectuals

William Petty “political arithmetic”

John Graunt: Natural and Politcal Observations made on Bills of Mortality (1662)

births and deaths = gains and losses. the same skills for tracking small margins of business was applied to deaths.

5. Royal Society of London (1661)
– centralized intellectual activity and skills
– “Nvllivs in Verba” no words or no speculation, only observation

Graunt is the convergence of these things
“The Table of Casualties” shows there is regularity of health phenomena
Was able to say that plague came from outside London, which was not the same for stable diseases
as a result, more quarantine enforced, and the plague stops affecting London

The Table of Casualties

Huygen’s adaptation of Graunt’s lifetable – median survival age
why asked this? how much should ask from people buying life insurance!

So how old is epidemiology? 1662 Graunt’s observation, our Big Bang. So 353 years in 2015!

Graunt_Observations

Marz. read the book, asks how to educate everyone in the world, as Morabia says we should. Need to promote “epidemiological literacy”. No one understands epi spontaneously; it requires learning to understand interpretation. That’s why epi should be taught in high school, and become part of general culture.
asks 16 year olds : what do you know about health?
marijuana does x. Can get STIs. HOW DO YOU KNOW?
Ah! That’s the question.

Things epidemiology taught me

Everything is about confounding.
Never leave home without a calculator.
Odds ratios are backwards and non-intuitive and misunderstood, but quite clever if used properly.

In other news, I plan on spending some time reading about the natural logarithm because I want to truly understand what the hell is going on in logistic regression, and how it can give risks/probabilities even when the odds ratio doesn’t approximate the risk ratio (?!).

In other other news, after months of research and asking around, I finally bought a little Opinel to fill the void left by the Victorinox of my childhood. Unassuming and efficient, it’s hard not to fall in love.