Book review: medical statistics
Today is a slow work day, so I finally finished reading my medical stats book.
When I put in my application for "development" funding from the Royal Academy of Engineering, one of the things I said I'd spend it on was a book on medical statistics. See, the statistics used by medics are really quite different from the statistics that I know. For a start, they're virtually 100% frequentist, whereas I was taught almost exclusively by Bayesians, who teach you really basic frequentist stuff, and then explain why (they think) it's Bad and Wrong. Be that as it may, when reading or writing medical papers, you have to be able to understand frequentist concepts, particularly P-values and confidence intervals.
So I got the money, and spent many happy hours on Amazon, and in various bookshops, looking for a book on medical statistics that would suit me. I settled on Essential Medical Statistics, as it presupposes a reasonable level of intelligence, has lots of worked examples, and has sections on how not to do it, which is a much-overlooked aspect of statistics, in my opinion.
Having effectively said "I need this book", and then been given it, I felt somewhat obliged to actually read it. Luckily, it is actually readable (and even has useful bits saying where you can skip a section or a derivation until you need it). Most of the chapters are about 10 pages, so I aimed for a chapter a day, on the understanding that there might be days, or even weeks, when I was too busy to read any. All in all, it took me about 3 months, which is not half bad. After reading each chapter, I made a one-line note to say what it was about, and how useful I thought it would be. So for example, Chapter 20 got a comment of "Extending logistic regression: could be useful for age modelling, data fusion and reading papers", so I can remember to review that chapter when I start doing age modelling.
Good things about the book:
Overall, I think this is a very good book, though. I'd certainly recommend it as an entry-level introduction for non-mathematicians.
When I put in my application for "development" funding from the Royal Academy of Engineering, one of the things I said I'd spend it on was a book on medical statistics. See, the statistics used by medics are really quite different from the statistics that I know. For a start, they're virtually 100% frequentist, whereas I was taught almost exclusively by Bayesians, who teach you really basic frequentist stuff, and then explain why (they think) it's Bad and Wrong. Be that as it may, when reading or writing medical papers, you have to be able to understand frequentist concepts, particularly P-values and confidence intervals.
So I got the money, and spent many happy hours on Amazon, and in various bookshops, looking for a book on medical statistics that would suit me. I settled on Essential Medical Statistics, as it presupposes a reasonable level of intelligence, has lots of worked examples, and has sections on how not to do it, which is a much-overlooked aspect of statistics, in my opinion.
Having effectively said "I need this book", and then been given it, I felt somewhat obliged to actually read it. Luckily, it is actually readable (and even has useful bits saying where you can skip a section or a derivation until you need it). Most of the chapters are about 10 pages, so I aimed for a chapter a day, on the understanding that there might be days, or even weeks, when I was too busy to read any. All in all, it took me about 3 months, which is not half bad. After reading each chapter, I made a one-line note to say what it was about, and how useful I thought it would be. So for example, Chapter 20 got a comment of "Extending logistic regression: could be useful for age modelling, data fusion and reading papers", so I can remember to review that chapter when I start doing age modelling.
Good things about the book:
- it doesn't assume too much background, but doesn't pander to innumerates - if you didn't get the concept first time round, you'll be expected to go back and revisit it rather than having it explained again. Exceptions are where there are common misunderstandings which might need beating out of people.
- It has enough maths, but not too much. For example, no derivations (yay!)
- If you do need more detail, there are plenty of good references
- Virtually every technique is demonstrated with worked examples
- You can either read it straight through (like I did) or dip in as needed
- It's very frequentist. Bayes is introduced in Chapter 33 (of 38) in very basic terms and never mentioned again. While I accept that most medical research doesn't use Bayes, it does have its uses, and I suspect it will become more common as time goes on.
- Some terminology annoyed me. For example, "log" always means natural logarithm, which is hard to remember as an engineer (we use "ln" for base e and "log" for base 10), but I think this might be a general mathematics convention that engineers break for convenience - like the i versus j argument. There were other instances though, like using multiple-character variable names - is pyar one value, or 4 multiplied together?
- It's so medicine-orientated that it's sometime hard to figure out how to apply a method when you're not dealing with "exposed" and "unexposed" people, or "healthy" and "disease" outcomes.
Overall, I think this is a very good book, though. I'd certainly recommend it as an entry-level introduction for non-mathematicians.