In my previous posts, I’ve written a lot about the importance of being in charge of your own care making informed decisions. Fortunately, the amount of information available to anyone with access to a computer and internet is nothing short of astounding. Unfortunately, this plenitude can also be discouraging for people who are unsure of how to find what they need, or interpret it. With this post, I’ll try to go into some detail about finding and reading information about diseases and treatments, including a brief overview of studies and statistical methods.
Websites like WebMD are a great resource for general medical information and help. It offers an easy to use search engine and helpful tools for very many patients. Furthermore, people can be fairly confident about the content as all the information is reviewed by an independent board consisting of four physicians. Basic sites such as these, however, are not well suited for more in-depth and ‘cutting edge’ research, or even basic epidemiological studies. For these, it’s likely that you’re going to need to look at some research papers. This is particularly important for a rare disease like carcinoid, where very little is known about the disease and most of the new information is released through journal publications PubMed is a very well organized database with articles related to medicine and biological sciences maintained by the United States National Library of Medicine at the National Institutes of Health. PubMed uses a specific system of organizing its content, partly due to the sheer number of articles. In order to search within PubMed effectively, knowing its MeSH vocabulary is invaluable. MeSH stands for Medical Subject Headings, and is used as a way to categorize all of the articles found within PubMed. A full overview would be difficult to do hear without the aid of visuals, but one can be found here: http://www.nlm.nih.gov/bsd/disted/video/. Pubmed will often have links to the full article, but even if they don't, should provide an abstract. Research and medical abstracts are a great way of getting a lot of information with a relatively small time investment, and shouldn't be taken for granted.
The importance of statistics is hard to underestimate in the modern scientific era. Because so much has to do with data, and statistics is the science of working with data, it is an integral part of almost all large scale endeavors. It is important to understand the advantages and disadvantages of working with information on a broad scale. In regards to things like medical interventions, it is invaluable to know the possible effect based on data from many individuals, and it is part of the reason that so called 'anecdotal evidence' is often viewed with such skepticism. Consider this in terms of simple games of chance, like that of flipping a coin. If a person were to hypothetically make all future assumptions about a coin based on three tosses, it is certain that they will be mistaken (because of the small and odd number of tosses). However, the more times a person flips a coin, the likelier it is that they will get a distribution of heads and tails that is closer to the 'true' probability of 1/2 for both. This is not dramatically out of line with what researchers attempt to do with clinical trials and observational studies, and it highlights the importance of having a large sample size. Testing a drug or trying to determine a risk factor for a certain condition is almost an entirely futile endeavor if done on a handful of individuals, particularly because unlike a coin toss, it is nearly always the case that there will be multiple factors in whether a person becomes sick with a certain condition, or recovers from one. Essentially, biostatistics is about taking what happens to a given group of people, and seeing what that means for a much larger group of people (referred to as the population).
At the same time, as important as it is to have a lot of data from many different sources, the more a statistical model says about a population, the less it may be able to say about an individual. Life, death, and disease are not a matter of coin flips, and it easy for people who know what the words 'median', 'mean', 'variance' and 'standard deviation' mean to be confused about what those measures actually say. A great example of this was written about by Dr. Steven J. Gould after his diagnosis of peritoneal mesothelioma. Initially despondent at discovering that the median life expectancy among those with his condition was 8 months, he was quickly relieved to find that his specific status meant he would likely live much longer, which he indeed did. Go back to the example about coin tosses. Imagine hypothetically that the odds of getting tails was still 1 in 2 overall, but the individual odds depended on what color shirt (please ignore the fact that this makes no sense) the person flipping the coin happened to be wearing. For example, a person with a red shirt has a 1 in 4 chance of getting tails, whereas a person wearing blue would have a 3 in 4 chance. Taken together, and flipped a sufficient number of times, the mean outcome will still be around 1 in 2 for heads or tails, but this doesn't necessarily say much for someone wearing a red shirt! This is useful to keep in mind the next time you're reading a study or abstract about a large population, and specifically lead you to ask how similar the people in the study are to you, in ways that are relevant (hint: for better or worse, socioeconomic status is almost always relevant).
Finally, there is one additional caveat to think about when reading studies based on statistical models. While it may seem counter-intuitive, statisticians pose their questions around the null hypothesis, or the hypothesis that there is no difference. For example, for two groups of people, one of which takes aspirin and the other a placebo, the null hypothesis would be that there is no difference between the two groups regarding some outcome, such as headaches. Conducting a study is an attempt at rejecting the null hypothesis, and not necessarily about proving that aspirin cures headaches. Strictly speaking, a single statistical study very rarely serves to prove anything, it allows the possibility that the null hypothesis can be rejected with varying levels of certainty. There are many reasons for this, but go back one last time to the coin toss. Say the coin was tossed 10,000 times and it landed 4,998 times on heads, and 5,002 times on tails. If you think about it, it doesn't really make sense to say that this has proven anything definitive, but it does make sense to reject the idea that either tails or heads will land significantly more times than the other. Of course, for practical purposes, we treat matters differently, but this proverbial coin has two sides too. While we may take it that aspirin relieves headaches if there is enough of a difference between groups, statistically significant differences may have no actual practical implications.
Here are some basic terms and ideas you will likely come across when reading research papers:
Test statistics: Examples include t-value, chi squared, and Fisher's exact test. These are the numbers resulting from the various appropriate statistical tests used in a study. They are used to obtain the more practical numbers that follow.
p-value: The probability of obtaining a certain test statistic like the one obtained by the study (for example the t-value). Practically speaking, it is the probability of the observed difference (say less people taking aspirin getting headaches) happening as a result of chance. So the usual cutoff of .05 indicates that there is only a 5% chance that the differences observed by the study could have come about assuming the null hypothesis is true (or there is no difference).
Confidence intervals: These are paired with a confidence level, such as 95%, and together give a range of likely values given what was observed by the study. For example, say a study found that the mean weight of adult males with some condition is 165, and gives a 95% confidence interval of 154 to 190. This indicates that there is a 95% chance that the 'true' mean, reflecting the entire relevant population and not just those in the study, lies between 154 and 190.
1-tailed vs 2-tailed: This has to do with direction of a study. If the study is 1-tailed, this means that the researcher is essentially assuming that the deviation will occur in one direction (for example, a supplement will only cause someone to lose weight, and not gain it). Because of this assumption, the p-value of a 1 tailed study will be smaller than one from a 2 tailed study, which simply predicts that there will be some difference, without saying how it will be different. This means that 2-tailed tests are "stronger", and 1 tailed tests should only be used if there is good reason to believe that the difference will only be in one direction.
Risk ratio/relative risk/rate ratio and odds ratio: These are two different things, but the interpretation can be treated relatively equally (keeping in mind that the former is likely to be a stronger indication). RR and OR discuss the impact of some variable on an outcome, and can be expressed as follows: A relative risk of 4 indicates that group had 4 times the risk of an outcome than a different group. The number can also be less than 1, for example, group A had .31 times the risk.
*All of the above information is available in "Basic & Clinical Biostatistics" by Beth Dawson.
In the next part, I will go over different types of studies, and the possible advantages and disadvantages of each.
Some links for your consideration:
Dr, Gould's article about median life expectancy
Painless Guide to Statistics from Bates College
Just for fun: A counter-intuitive probability game called the Monty Hall Problem.
Test statistics: Examples include t-value, chi squared, and Fisher's exact test. These are the numbers resulting from the various appropriate statistical tests used in a study. They are used to obtain the more practical numbers that follow.
p-value: The probability of obtaining a certain test statistic like the one obtained by the study (for example the t-value). Practically speaking, it is the probability of the observed difference (say less people taking aspirin getting headaches) happening as a result of chance. So the usual cutoff of .05 indicates that there is only a 5% chance that the differences observed by the study could have come about assuming the null hypothesis is true (or there is no difference).
Confidence intervals: These are paired with a confidence level, such as 95%, and together give a range of likely values given what was observed by the study. For example, say a study found that the mean weight of adult males with some condition is 165, and gives a 95% confidence interval of 154 to 190. This indicates that there is a 95% chance that the 'true' mean, reflecting the entire relevant population and not just those in the study, lies between 154 and 190.
1-tailed vs 2-tailed: This has to do with direction of a study. If the study is 1-tailed, this means that the researcher is essentially assuming that the deviation will occur in one direction (for example, a supplement will only cause someone to lose weight, and not gain it). Because of this assumption, the p-value of a 1 tailed study will be smaller than one from a 2 tailed study, which simply predicts that there will be some difference, without saying how it will be different. This means that 2-tailed tests are "stronger", and 1 tailed tests should only be used if there is good reason to believe that the difference will only be in one direction.
Risk ratio/relative risk/rate ratio and odds ratio: These are two different things, but the interpretation can be treated relatively equally (keeping in mind that the former is likely to be a stronger indication). RR and OR discuss the impact of some variable on an outcome, and can be expressed as follows: A relative risk of 4 indicates that group had 4 times the risk of an outcome than a different group. The number can also be less than 1, for example, group A had .31 times the risk.
*All of the above information is available in "Basic & Clinical Biostatistics" by Beth Dawson.
In the next part, I will go over different types of studies, and the possible advantages and disadvantages of each.
Some links for your consideration:
Dr, Gould's article about median life expectancy
Painless Guide to Statistics from Bates College
Just for fun: A counter-intuitive probability game called the Monty Hall Problem.
No comments:
Post a Comment