Apparently not. The divorce rate is the lowest it's been since 1970. But you wouldn't get that impression from reading some recent articles.
This opinion piece in the New York Times by two faculty members at Wharton lay out the problems with the recent analysis of a chances of making it to 25 years. They write...
"The census data come from a survey conducted in mid-2004, and at that time, it had not yet been 25 years since the wedding day of around 1 in 10 of those whose marriages they surveyed. And if your wedding was in late 1979, it was simply impossible to have celebrated a 25th anniversary when asked about your marriage in mid-2004.
If the census survey had been conducted six months later, it would have found that a majority of those married in the second half of 1979 were happily moving into their 26th year of marriage. Once these marriages are added to the mix, it turns out that a majority of couples who tied the knot from 1975 to 1979 — about 53 percent — reached their silver anniversary.
This surveying glitch affected only the most recent data. Still, factoring in an appropriate adjustment yields the conclusion that divorce rates have been falling, not rising. This is not just statistical smoke and mirrors"
Wow! The data could have been presented with a lot more nuance. It's a shame the original story was so wrong and people probably won't here about the correction.
Saturday, September 29, 2007
Saturday, September 22, 2007
Quit While You're Behind
When should a medical study of a drug/therapy be abandoned? Merck has announced that a trial of a candidate HIV vaccine was stopped early. Their original plan was to continue through 2008 and enroll 3,000 HIV negative volunteers. Instead, the study has stopped after enrolling just over 1,500 participants. Why?
Because of discouraging results. The study found that about 3% of subjects became infected with HIV with slightly higher rate among those who got the vaccine compared to those who got placebo injections.
The decision to cut a study short has to weigh complex, competing considerations. If the vaccine is truly a dud, it is unethical to continue to expose new volunteers to the burdens and risks of participating in the study -- the money and resources could be better deployed to test other promising HIV prevention stratgies (see disclosure). But are we sure it doesn't work? We'd want to be sure.. After all, formidable resources have been invested to answer the question and scientists owe the public a conclusive answer.
Statisticians look at the data in hand and ask "If the study was completed, what the chance the study would favor the vaccine?" This calculation depends on how far into the study we are and projections about how the vaccine will look in the (yet uncollected) future data.
The study was planned to continue until there were around 90 HIV infections (a grim reality). Based on this, the vaccine would be proved effective if 35 (or fewer) of the 90 infections occured among the vaccinated. Once there are more than 35 HIV+ among the vaccinated, the vaccine will be condsider a failure. At the half-way point, there were 24 HIV+ subjects among the vaccined.
How could the vaccine to come back from that bad start? Only if among the next 45 infections, fewer than 11 of those infections (35 - 24) occur in the vaccinated group - a split of 11 v. 34. How likely is that?
Let's assume the vaccine is very effective (even though the current data contradicts that) and it cuts the risk of HIV infection by 50%.
Adopting this optimism, we can calculate the change of a "positive" trial using the binomial distribution and get an approximate probability of 0.13. Thus, even if the vaccine worked, the odds are more than 6:1 against a positive result. Unlikely. Even someone who believes the vaccine works would concede, given the available data, the vaccine is going to "fail". I can post my calculations later for the technically inclined.
Decisions about whether a trial should stop or continue are made by an independent board of experts who consider numerous aspects of the data. For instance, the count of infections was 19 vaccine v. 11 placebo among those who'd received more than 1 injection. This raises a theoretical, if not conclusive, possibility of increased risk of HIV from the vaccine. This is discussed an article in the San Francisco Chronicle.
Disclosures I am the biostatistician for a trial of a non-vaccine HIV prevention strategy -- pre-exposure prophylaxis.
Because of discouraging results. The study found that about 3% of subjects became infected with HIV with slightly higher rate among those who got the vaccine compared to those who got placebo injections.
The decision to cut a study short has to weigh complex, competing considerations. If the vaccine is truly a dud, it is unethical to continue to expose new volunteers to the burdens and risks of participating in the study -- the money and resources could be better deployed to test other promising HIV prevention stratgies (see disclosure). But are we sure it doesn't work? We'd want to be sure.. After all, formidable resources have been invested to answer the question and scientists owe the public a conclusive answer.
Statisticians look at the data in hand and ask "If the study was completed, what the chance the study would favor the vaccine?" This calculation depends on how far into the study we are and projections about how the vaccine will look in the (yet uncollected) future data.
The study was planned to continue until there were around 90 HIV infections (a grim reality). Based on this, the vaccine would be proved effective if 35 (or fewer) of the 90 infections occured among the vaccinated. Once there are more than 35 HIV+ among the vaccinated, the vaccine will be condsider a failure. At the half-way point, there were 24 HIV+ subjects among the vaccined.
How could the vaccine to come back from that bad start? Only if among the next 45 infections, fewer than 11 of those infections (35 - 24) occur in the vaccinated group - a split of 11 v. 34. How likely is that?
Let's assume the vaccine is very effective (even though the current data contradicts that) and it cuts the risk of HIV infection by 50%.
Adopting this optimism, we can calculate the change of a "positive" trial using the binomial distribution and get an approximate probability of 0.13. Thus, even if the vaccine worked, the odds are more than 6:1 against a positive result. Unlikely. Even someone who believes the vaccine works would concede, given the available data, the vaccine is going to "fail". I can post my calculations later for the technically inclined.
Decisions about whether a trial should stop or continue are made by an independent board of experts who consider numerous aspects of the data. For instance, the count of infections was 19 vaccine v. 11 placebo among those who'd received more than 1 injection. This raises a theoretical, if not conclusive, possibility of increased risk of HIV from the vaccine. This is discussed an article in the San Francisco Chronicle.
Disclosures I am the biostatistician for a trial of a non-vaccine HIV prevention strategy -- pre-exposure prophylaxis.
Subscribe to:
Posts (Atom)
