Many thanks to the people who responded to my query (bottom of email). Summarises follow: Particular thanks to Dr. Alex McMahon, whose paper brought to my attention propensity score matching: ' Matching in drug epidemiology is usually done by 'propensity scores'. You don't need a special package as it's easy to do in a normal stats package. Once you have created the scores you can box them into quintiles thus making the matching easier than having to do individual matching on lots of variables. See the references in my paper of last year: Approaches to combat with confounding by indication in observational studies of intended drug effects. McMahon AD. Pharmacoepidemiology and Drug Safety 2003; 12; 7: 551-558. ' Thanks to Professor Mike Campbell: ' To my mind logistic regression should largely be confined to situations where one can only have one event, eg death. Poisson regression is really only valid when the counts are truly independent, eg numbers of deaths in an age/sex group. When you have counts within a unit, such as number of recurrances, then you should be using either a negative binomial or an overdispersed Poisson, or some form of random effects or multi-level model. ' Thanks to Neil Hawkins: ' Couple of quick suggestions: 1) include the matching variables as covariates in your analysis rather than try to match on them. Alternately you might stratify your analysis by the matching variables. 2) The decision whether to analyse the number of events or the number of patients suffering one or more event should be primarily driven by the question you are addressing. If the question is one of resource consumption, then you are probably interested in the number of events occurring. If patients on average suffer more than one event, but you still analyse the number of patients suffering from events, then you will under-estimate resource consumption accordingly and need to consider if this will compromise any inference you draw from the analysis. ' Thannks to Dr. Peter Flom: ' You might want to look at modelling multiple events in a survival analysis context. This is covered in Therneau & Grambsch (2000) Modelling Survival Data: Extending the Cos Model. pub by Springer. They use R and Splus quite a bit. I believe that it's also discussed (on a less technical level) in some of Paul Allison's work, but that mostly uses SAS ' Original query: I'm writing up a SAP for a prospective observational study involving health resource outcomes of two competing drugs. My client wants me to put in far too many primary pieces of analysis but that's just how it is. I have two questions - any help on either much appreciated - I'll summarise and post back the responses. I've been asked to conduct patient matched analysis (i.e. matching 1 patient on drug X with 1 on Y) - I've been given the variables to match on. My background is econometrics so I don't know much about this topic. Is there a package out there anyone can recommend that can do the matching in either R or Stata or Ox - I've not a clue as to what to do if 2+ people from drug X perfectly match 1 person on drug Y and the fact my sample won't be equal sizes - likely to be 2 to 1. The other analysis is using standard technques to model the occurances of certain outcomes (eg hospital visits) in a specified period. I'm doing the usual time to event analysis, but the client also wants me to model either the number of occurances in the period (Posisson/Neg Binomial model) or just whether the event occurs (Logistic). My question is there a guide as what percent of patients having a count greater than 1 is a cut off that dictates whether to model the counts or just whether the event occurred? I want to put this cutoff in my SAP so I can use it dictate my modelling - no idea if this is legitimate? Any help much appreciated. DISCLAIMER: The information in this message is confidential and may be legally privileged. It is intended solely for the addressee. Access to this message by anyone else is unauthorised. If you are not the intended recipient, any disclosure, copying, or distribution of the message, or any action or omission taken by you in reliance on it, is prohibited and may be unlawful. Please immediately contact the sender if you have received this message in error. Thank you.