That doctors and the public are misled by unpublished drug trials
I worked in the drug industry for 20 years conducting Phase II and III (Premarketing) trialson new drugs. The data from these trials were submitted to the national drug licensing agencies (e.g. FDA in the USA). It was incumbent upon us to submit ALL data from these trials and serious efforts were made to do so. It is on the basis of these data, NOT the published data, that a drug is given a licence for marketing or not. If a drug comapanyfails to submit all the data it has, it is obviously culpable and may have committed fraud (though given the amount of data, human error is also possible - usually several large trucks were required to carry the volumes of paper to the regulators.)
The problem of publication bias is another problem altogether and it is serious for the reasons Ben Goldacre gives. But it's probably true that most doctors are less 'nerdy' than him and don't feel the need to personally examine all the trial results, being content to leave this task to the agencies set up for the purpose. I ran several nagative trials and was keen to get them published. I provided the results of our analyses to the investigators, they had their own copies of the original data, but few were interested in doing the work necessary to produce a publication because they knew that the chances of a journal accepting it were very low. THis problem was compounded by another factor that Dr Goldacre does not mention. Most of these trials were multicentre studies where many doctors co-operated on the same study protocol in order to obtain sufficient patients to achieve a statistically reliable result. Doctors need publications to pad out their CVs and get ahead in their careers. So, oftentimes, doctors wrote up and published the results obtained in their own small subset of patients, especially if that subset happened to give a positive result in the context of an overall negative study. I couldn't stop this.