the Seattle epidemiology research and Information Center in collaboration with the Department of Veterans Affairs the VA employee education system and the University of Washington Department of Epidemiology present the 2002 VA epidemiology summer session okay welcome back to general biostatistics the topic will now be looking at is common statistical tests that are used in comparing proportions of two groups so in the last lecture we were looking at comparing means of two groups now we’ll look at comparing proportions so remember that when we compared means we had a continuous outcome of interest now what we’re interested in our groups that have a dichotomous outcome of interest so these are data that we typically can see displayed in a two by two table we’re going to review some methods for summarizing information in two by two tables we’re going to look at how to test for differences in proportions between two groups using the chi-squared statistic and we’ll see that there’s an underlying methodology for this but also a shortcut formula for a two by two table chi-square statistic so as I said the outcome of interest is now a dichotomous one we’re looking at the proportion of of individuals that have the disease or an event of interest and outcome of interest that can be expressed as success or failure diseased or non diseased response non-response question that I have have received is why would be sometimes be interested in looking at proportions rather than always looking at differences in means so for instance suppose we were interested in response between two groups we might actually have more information if we compared the differences in mean response times between groups but sometimes we might not have the actual times we might just have the information on whether by a certain point in time say six months there had been a response or non-response so what we have is a situation in which we’re interested in whether two groups are the same with respect to some dichotomous outcome we can look at displaying this information in a two-by-two contingency table so here suppose I use the example of having two treatment groups a and B so here we have 50 observations in treatment group a and we have 70 observations in treatment group B so think of these as 50 patients treated by treatment a and 70 treated by treatment B and suppose that by six months we were looking to see whether they responded to treatment yes or no so what we have then are the totals the total number of responses was 50 for non responses was 66 totaling the hundred and twenty which equaled the fifty patients treated by treatment a and seventy treated by treatment B now if we were interested in whether they differed with respect to response we could look at the and calculate the proportion responding under treatment a and treatment B so the proportion responding takes the total who responded in treatment a 37 divided by 50 which is 74 percent or a proportion of 0.74 whereas in treatment group B 17 out of 70 or 0.2 for responded does there appear to be a difference in response between the two treatment groups yes and in fact it appears that there’s a much higher response in treatment a than in treatment B the question is is this difference in response likely to be one that we’d get just due to chance alone if there was truly no difference in response rate between the two treatments so we see we can calculate proportion responding in both tree mence what’s the overall proportion responding 54 divided by 120 which is 0.45 or 45% so how could we summarize this information we just talked about I’m using 74% as a percentage responding in a and 24% responding and B there are other ways that sometimes can summarize this information we could say based on this there’s a 50% absolute difference

in response rate between treatment a and B where there’s a higher response in a we could look at the relative difference in response rate compared to one of the treatments and so here I could look at the difference between a and B relative to B so if I took that difference of 0.5 divided it by the response rate in treatment B of 0.2 for I would see that I would get a ratio of 2.08 and if I multiplied by the hundred that would say that the the proportional increase in response rate was 208 percent higher in a relative to B so that you may see it reported as an absolute difference a relative or proportional difference you might even see it given by the odds of responding so what are the what’s the definition of odds the definition of odds is it’s the probability of responding divided by the probability of not responding so in treatment a the probability of responding was 37 out of 50 what was the probability of not responding 13 out of 50 let me take you back to the table so 37 out of 50 responded 13 out of 50 did not respond and by definition we could consider the odds to be two point eight five so the odds of response in treatment group a is two point eight five similarly we could look at the odds of respond in treatment B by taking the probability of responding which was 17 out of 70 divided by the probability of not responding and we would get 0.32 a common summary might be the odds ratio of response so it compares the odds in treatment a to the odds and treatment B so we would take 2.8 5 divided by 0.32 and we would say that the odds of responding are nine times higher in a than in B so all of these are just different ways of summarizing the same information what would you conclude regarding any of these regarding response rates in in the two treatments Haizi a seems to be better in terms of response rate than B so in this statistical comparison that we’re interested in making we’re looking at the question is the difference in the response rate that we observed bigger or larger than what we would expect due to chance alone if there was truly no difference between the two treatments so let me take you through reasoning how you might how we use the chi-square statistic to calculate this so here’s that same 2×2 table but we’re missing the inner cells so what we know is there were 50 people treated by treatment a 70 by B and that the overall response number of responses was 54 and 66 did not respond okay so I just have the marginal totals of this table well by reasoning through this what frequencies would we expect if the response rate was the same by treatment what would you expect say that again please so you’d expect 54 divided by 120 which is 45% is that what you said so so if we looked at just the total who responded which is 54 divided by 120 remember that was 45 and so if the response rate was the same in both treatment groups wouldn’t you expect 45% of the 50 people treated by treatment a to respond and when do you expect 45% of the 70 treated by treatment B to respond yes so let’s fill out the rest of the table so 45 percent of 50 is twenty two point five forty five percent of seventy is thirty one point five these are the numbers we would have expected now you’re going to tell me that we can’t have half of a person responding but this is just for the purposes of looking at expected counts and how would we fill in the number that don’t respond right subtract from the total so 15 minus twenty two point five gives the remainder the number we’d expect not to respond 70 minus thirty one point five gives the number we’d expect to not respond in treatment B and so here they are just filled in so now this is a table these

are the expected frequencies of response and non-response if there was truly no difference how does that compare to what we observed that’s the next table so can anybody tell me what they observe here’s a comparison this is what we observed in our treatment groups here’s what we’d expect if there was truly no difference in response rate how do these two tables compare they’re much different so what we see is that there appears to be a higher number who respond in treatment a than what we’d expect right and a lesser number who don’t respond in treatment a now compared to what we expect there’s a lesser number who don’t respond in treatment B so what the chi-square statistic does is it makes a comparison of the observed to the expected counts so we could eyeball this and we say they look much different we’d calculate the chi-square test statistic to make a formal comparison of observed to expected counts and we would do this for all four cells of this table so the chi-square statistic is one that looks at a comparison of observed and expected it squares it because some of those comparisons might be negative some might be positive so we’re looking at a squared difference divided by expected now suppose that what we observed in our data was what we would have expected then how would this guy what would this numerator look like zero so if what we observed was exactly what we expected the observed and expected counts were the same this numerator would just be equal to zero what would the value of the chi-square statistic be zero if what we observe is quite different than what we’d expect what happens to the numerator it increases so as the difference between observed and expected increases the value of the chi-square statistic becomes larger and larger so we’re going to use that as a test then by calculating the test statistic we can use this as a test of Independence or another way of saying this is that we can test that there’s no association between response and treatment another way of saying it is that the two groups the two treatment groups are the same or homogeneous with respect to treatment response so we can use this test statistic which follows a chi squared distribution it’s another distribution it’s not bell-shaped like the normal or the T but it is a distribution that depends on the degrees of freedom so how would we calculate the test statistic based on this data what we saw was the observed counts that are always given in the first place the expected counts that we had from that 2×2 table how these were expected based on what right this was based on expecting 45% to respond in both treatment groups in other words that there was no difference in response rate in the treatment groups so if we look at observed minus expected squared divided by expected total these for all four cells we end up getting a chi-square of 29.1 with one degree of freedom is that close to zero no not close to zero is it likely that we would have observed these differences just due to chance alone well we don’t know yet until we look at the appropriate chi-square distribution and the probability associated with it based on its appropriate degrees of freedom so for our course all that we’re going to worry about is a chi-square with one degree of freedom so I don’t have it listed here but it’s in the chi-square table at the back of the textbook when we have a two-by-two table the degrees of freedom are always the number of rows minus one times the number of columns minus one how many rows do we have – how many columns to rows minus one is one columns minus one is one degrees of freedom is one and the only value with a chi-square with one degree of freedom that we have to worry about is that the probability or significance level of 0.05 is associated with a chi-squared

statistic with one degree of freedom equal to 3.84 so how does 29.1 compare 23.84 much higher indicating that the p-value is going to be something much less than 0.05 does that make sense what’s the whole point of this chi-square statistic it’s just looking at how much discrepancy there is between observed and expected how much discrepancy there is between thirty seven and twenty two point five between thirteen and twenty to twenty seven point five between 17 and 30 1.5 etc and it’s reflected then in the sky square statistic of twenty nine point one which is very large if there was little disagreement between observed and expected this Chi square statistic would take on values closer to zero so again reiterating if what you observe is what you expect the numerator would be equal to zero this Chi square statistic would be low or equal to zero okay so this was the long way of calculating the Chi square statistic there’s a shortcut for two by two tables that takes you don’t have to calculate expected counts at all all that you have to do is label the cells as a B C and D then we can talk about the row totals we can talk about the column totals and the shortcut formula says take the overall total of 120 and then multiply it by the cross product of a times D minus B times C so in your notes make sure that says C a times D minus B times C squared divided by the product of the row and column totals so just taking the row totals the column totals of 54 times 66 times 50 times 70 so that’s on the next slide which shows we take the overall total and then just take a times D minus B times C that quantity squared divided by the product of the column times the row totals and we’ve come up with the same custom of 29.1 all that it is is a shortcut if you’re doing it by hand now I can show you this in Stata as well and what were the four cells a b c and d 37b was 13 then I’m going to indicate I’m going on to the next role with a backslash 17 and 53 so using the tab I command it’ll give me the 2 by 2 I’m going to put a comma there and indicate that I would like the chi-square statistic and I’d also like it to show me some roll percentages so I’d made a mistake thank you okay so now here’s that 2×2 table that you have in the notes so what is this showing says drug a drug B 74 percent meaning 37 of 50 responded 17 out of 70 or 24 percent responded in B and what we see is the chi-square statistic of 29.1 that basically has a p-value that’s very close to zero very close to zero now some of you have said well this is one way the chi-square test is one way there is another way to do a Z test with proportions and the Z test with proportions as is just as we had a difference in means with two groups we can look at a difference of proportions with two groups and in Stata we would use something that’s called proportion test PR test again immediate so I’ll tell it that I had 50 in treatment a and the response proportion was 0.7 for and then there were how many and treatment B 70 remember with a response of 0.2 for so now here’s a two sample test of proportions we get AZ test of Z which is 5.4 what’s the p-value same thing that we got when we looked at the chi-square in fact there’s a an algebraic relationship that if we squared this Z of 5.4 we would get that chi-squared value of 29 that we calculated by hand it’s just an

algebraic relationship the bottom line is that our interpretation is the same whether we’ve calculated the chi-square test or a Z test that was looking at the difference in these two proportions so we will not go into the other methods outside of the chi-square test for looking at difference in proportions but what if I looked at the 95% confidence intervals for each of these proportions they’re calculated in the same way based on this proportion of 0.74 and placing an interval around it so we see the first group treatment a ranges from about 0.6 up 2.86 the confidence interval for treatment b ranges from about 0.1 3 up to 0.3 for similar or different very different well i don’t think they overlap they go from 61.8% to 86.1% for treatment a the 95% confidence interval for treatment B goes from about thirteen point nine percent up to 34 percent they’re very distinct so this would also tell us that there appears to be a very statistically significant difference in the proportion of response between the two groups so I wanted you just to be able to see that although we’re just although we’re emphasizing the chi-squared as one way of easily looking at a test of whether two proportions are different between two groups there are other methods we could do like we just did a Z test of the difference in proportions when the cell counts are small something known as Fisher’s exact test could be used we also Illustrated that we could look at the separate confidence intervals for a and B but we could also look at the confidence interval for the true difference in proportions between the two groups and then these can be expanded to looking at odds ratios and confidence intervals or even ways of combining information from multiple tables but these are all advanced top we won’t be looking at today or ever so well we will be focusing just on just on the chi-square as a simple test of the differences in proportions so so I hope that this made some sense if we looked at testing differences between two proportions we’re really asking are the groups the same with respect to some outcome or characteristic another way of phrasing that is by asking the question is there an association between group and characteristic in other words between treatment group and response another way of phrasing that is asking our group treatment group and response independent but we would use a chi-square statistic to answer the question however it’s phrased there is a question the degree of freedom you had in the last kind square again or how you write so for our purposes when we have two groups and we’re comparing proportions the degrees of freedom are always 1 but the way we would do this for any contingency table that has our rows and C columns is that the degrees of freedom is the number of rows minus 1 times the number of columns minus 1 that’s the general formulation for degrees of freedom let’s see another question is the use of the chi-square statistic valid when you have more than two groups yes so if you have three groups and we will be talking about that tomorrow if you had three groups and you were interested in knowing is the percent responding and treatment group a B and C all the same we could use a chi-square statistic that would have three rows and two columns so we’ll be covering that as well in the next lecture let’s review last night’s homework remember that our topic yesterday was looking at setting the framework for statistical inference by looking at sampling distributions remember that sampling distributions are theoretical distributions but we can think of it as the probability distribution that describes the values that different sample statistics may take and under certain assumptions the values that that sample statistic may take will be approximately normally distributed centered at a mean so we can think of this as the mean of the statistic with a standard deviation which reflects the variability or spread in the statistic it’s this Sigma the Sigma that’s associated with the values of the

statistic that’s referred to as the standard error so two of the most common sampling distributions that we focused on were the sampling distribution of the sample mean x-bar so if we place this into the context we can think of in theory if we sampled from a population that had a true mean mu and a standard deviation Sigma and if we could compute all the possible sample means from all possible samples of size little n and plot them on a histogram the shape of that histogram would be approximately normally distributed the mean of the sample means would be the same as the population mean mu and the spread or the standard deviation of the sample means would be depend on the standard deviation in the population Sigma as well as the sample size so the spread in the sample means as Sigma over the square root of N and similarly if we had a dichotomous outcome the theoretical sampling distribution of all possible sample statistics has a mean that’s equal to the true population proportion and a spread that’s equal to the square root of the true population proportion P times 1 minus that or Q divided by the sample size n so the reason we spent some time on this is that these sampling distributions underlie the basis for statistical inference both in terms of estimation as well as hypothesis testing so keep in mind that with estimation for a confidence interval estimate we’re taking the sample statistic that we observed based on our single data set forming an interval around it by adding and subtracting an appropriate Z or T tabled value depending on the alpha of interest times the standard error that corresponds to that theoretical sampling distribution of the sample statistic at hand and in general for hypothesis testing the test statistic that we use in a hypothesis test is formed by taking the value of our observed sample statistic from our data minus the hypothesized value divided by the standard error of the sample statistic so we’re using the spread in the sample statistic or the standard error both in forming the confidence interval as well as in setting up the test statistic for the hypothesis test so let’s I use this theoretical framework then and look at a few examples from the book one example is one in which you’re given hypertension prevalence estimates based on the N Haynes survey conducted from 1976 to 1980 and it’s look we’re looking at differences here in hypertension prevalence between whites and blacks so the question at hand is is there a difference in the prevalence of hypertension by race so the fact that these are estimates means that based on our data we have a dichotomous outcome an individual either has hypertension or not based on a certain cutoff so we have two estimates here in the book it’s expressed as a percentage twenty-five point three percent of whites and thirty eight point six percent of blacks and the standard error that corresponds with it is also a percentage 0.9 percent is the standard error in other words that’s the variability in the sample statistic the square root of P times Q over N of 0.9 percent in whites and 1.8 percent in blacks now in order to work with these proportions since these were expressed in percentages all that we need to do is divide by a hundred so we would see the sample proportion of whites is 0.25 three the sample proportion of blacks with hypertension is 0.38 six and the standard errors as well would need to be divided by a hundred for a standard error of 0.009 and whites and 0.018 in blacks so how could we construct confidence intervals for the true but unknown prevalence of hypertension in whites and in blacks so for each group we would take the sample proportion form an interval around it by adding and subtracting 1.96 corresponding to an alpha of 0.05 for a 95% confidence interval times the standard error that’s based on the square root of P hat Q hat over N now keep in mind that what we were given in this example was just the standard error so we were already given

the square root of P hat Q hat over n do we need the sample size no because we’re already given the calculated standard error so for whites this confidence interval could be expressed as the sample proportion of 0.25 3 plus or minus 1.96 times the standard error of 0.09 and if we go through the math what we see is that we have a 95% confidence interval ranging from point 2 3 5 to 0.271 this is a confidence interval for the true population proportion this allows me to say that I’m 95% confident that the true proportion of of whites hypertension in this time period is contained in this interval that ranges between 23.5% up to 27.1% similarly for blacks we could calculate the 95% confidence interval by taking the estimated proportion that we got of 0.38 6 adding and subtracting 1.96 times the standard error of 0.01 8 and again going through the math I would be able to say that I’m 95% confident that this interval that ranges from 35 point 1 up to 42 point one percent contains the true population prevalence of hypertension in blacks so now how would we interpret these they’re both fairly narrow intervals is there a statistically significant difference in the proportion with hypertension in White’s versus blacks yes because we see a fairly narrow confidence intervals that don’t overlap so by looking at this confidence interval we see that there appears to be a higher prevalence of hypertension in blacks as opposed to the prevalence in whites now just a few questions that we could also ask these are 95% confidence intervals what would happen if we were asked to construct 99% confidence intervals the only change is that instead of using a Z of 1.96 we would use a Z of 2.5 seven we have now 99% confidence but what happens to the width of these intervals they increase whereas if we were looking for a 90% confidence interval in other words one in which we have less confidence the Z statistic would be 1.645 and we would have much narrower intervals so again remember these Z’s of 1.96 two point five seven and one point six four just correspond to the areas in two tails of the normal distribution that we can find in the tabled values now that example that we had looked at was looking at proportions estimated proportions and calculating confidence intervals for the truth and the populations the broader populations of whites and blacks suppose that we look at this next example which is dealing with a continuous measurement that of body mass index measured by taking weight in kilograms divided by height and meters squared in this scenario we are presented with a study of 58 middle-aged men who develop later in life diabetes and were told that in the sample of 58 men the sample mean BMI was 25 kilograms per meter squared the sample standard deviation was 2.7 kilograms per meter squared and supposed that we were interested in constructing then a 95% confidence interval for the true mean BMI in the population of all middle-aged men who develop diabetes we take the sample mean again sample statistic of the sample mean form an interval around it by adding and subtracting now a tee times the standard error of the mean we don’t know the true population standard deviation of BMI we get our best estimate by using the sample standard deviation s so the standard error is given by s over the square root of n and because of our lack of knowledge of Sigma we would use a tee still with alpha of 0.05 split in two tails but the degrees of freedom depend on the sample size minus one or 57 so just to follow through on this we would take the sample mean of 25 plus or minus the tabled value of two times the standard error of the mean that we get by taking 2.7 divided by the square root of 50 and our confidence interval range is then from twenty four point three kilograms per meter squared up to twenty five point seven kilograms per meter

squared what does this allow me to conclude I can say I’m 95% confident that this interval contains the true population mean BMI for middle-aged men who develop diabetes I can also say that values ranging from 24 point 3 up through twenty five point seven are consistent with the truth based on this sample of twenty of fifty eight middle-aged men good day my name is Gail Reiber I’m the director of the VA summer epidemiology program here in Seattle I’m a VA career scientist and a professor at the University of Washington in Health Sciences and epidemiology during the summer program we brought some very interesting faculty and to address topics of interest to our VA colleagues I’m happy to welcome dr. Mike Gaziano dr. Gaziano is the director of maverick the Massachusetts and veterans epidemiology research and Information Center dr. Gaziano is also a professor at Harvard University he teaches grant writing at the School of Public Health so I would like to ask dr. Caggiano some questions that might be helpful to future researchers who are going to be writing grants so if grant writing is in someone in the audience’s future what advice do you have in terms of getting started there’s a lot that has to happen before you even begin to write the grant the first thing is you have to have a good idea good ideas can come from anywhere it can come from your clinical practice it can come from an RFA it can come from prior research that you’ve done the next step is to search the literature and make sure your idea is at least somewhat novel and that should be followed by a discussion of what funding agencies might be receptive to your idea these can include the VA NIH private foundations specialty societies or even the university it’s essential to have a team assembled before you begin writing this could include a statistician collaborators budget analysts to help put your budget together and then it’s time to begin to write okay grant seem to be fairly complex and there are a lot of different portions of grants what are actually the most important portions of grants and how would one go about thinking through those portions well the grants are structured somewhat differently depending on the funding agency and the first step is to understand the rules of the grant application for each agency but most grants have four key components specific aims a background and significant section a section for previous work and your methods section the methods is the meet of the grant the specific aims is designed to lay out your objection your objectives and your hypotheses and to give an inkling and how you’re going to implement your aims the background and significance defines the problem and identifies the gaps in knowledge that your study will answer and alludes to how your study will fill in those gaps the significance portion of this part of the grant tells why it’s important to know that information because it’s something that that we is essential an essential piece of information to reduce morbidity or mortality or save dollars because the next section is is previous work the previous work section derives from work of others in the field or from your own preliminary work it’s a great place to put pilot information and what they previous work section is designed to do is to give them the reviewer of your grant an idea of the background behind your methods the scientific method behind your methods the scientific underpinnings of the elements of your methods that you’re going to choose to implement your study but the real meat of us of the grant application is the method section and it has several components I begin with a an overview of the methods section telling the author what’s going to be laid out in the method section a simple paragraph defining your overall design is essential then the next steps in the in a method section depends a bit on your design you usually define your pot study population inclusion and exclusion criteria you want to define the measures that you’re going to employ to collect data your baseline instruments lab measurements that you might use a plan to follow up the participants if it’s a observational study you want to clearly define the exposures that you’re interested in if it’s a trial you’re going to define the interventions

fundamental to any method section is how you’re going to deal with the data once you’ve collected it so a careful analysis plan that’s laid out with the help of a statistician is essential also a statistician can help define the power that your study is going to have to answer the specific questions and last I think it’s important to bring up issues about limitations in your in your grant but don’t bring up fatal limitations without having some way to adjust those problems you should have an answer for the limitations that you raise okay the abstract seems to be pretty important to people on review panels describe the importance of the abstract from your perspective the abstract is key and I always write the abstract at the very end of the process and it should summarize the your aims your study population the problem the caps in the information and then your basic approach the reason an abstract is so important is because it sets the tone for the reviewer there are many reviewers on a study section that will review the abstract and because they’re not the primary reviewer the abstract will be a fundamental area where they’re going to get a lot of information to be able to review your grant and to do a focus review so that the abstract is very important and should should not be done in haste at the last minute okay you’ve mentioned that the people that review the grant are going to have a lot to say about things would you describe what’s going to happen to this grant once it’s submitted to an agency well typically the grant first undergoes some sort of administrative review where people who are not concerned about the science will see if you follow the rules and grants can be rejected if you don’t follow the rules you didn’t follow the font-size regulations or the number of pages in the grant your grant can be sent back sent back without a scientific review it makes that first hurdle it gets passed on to a review panel often called a study section which is comprised of your peers they’re usually people who are some expertise have some expertise in the particular field in which your your grant falls the study section will score the grant and study sections from various agencies have slightly different strategies usually there’s a numeric score in addition there is a careful critique of the grant that will be provided to the the applicant the scoring process allows that study section to rank all the grants that were reviewed at that particular point in time to rank them from top to bottom and often more important than the numeric score is the percentile score the next step is those scores are passed back to the funding agency because study sections are often independent of the funding agency and it’s up to the funding agency to decide what whether there’s enough funds to fund your grant and typically as an NIH or VA study section will fund twenty to thirty percent of the applications in one given study section so if we don’t do well the first go-around what are our choices well there are several options the the the you could receive a fundable score which means you have to get to work because you’ve made a commitment grant is a contractual arrangement with the funding agency and if you ever want to get another grant from that agency you better deliver on the promises made many grants don’t get funded the first time and what you have to do is do a careful review of what the problems in the grant were so a careful review of the of the critiques you often will get the critiques before you get a funding score before you get a funding decision and that’ll give you a clue as well as to whether or not this will be a fundable grant if it’s not funded it’s often worth having a discussion with the funding agency to see if it’s worth three submitting many times it is most grants are not funded the first time around and persistence does pay off sometimes your idea just wasn’t ready for the study section you have to educate the study section before they’re willing to accept your idea as as a viable path sometimes when grants aren’t funded and it’s a new investigator just starting out would you recommend broadening the portfolio and looking to different sources that may be not be merit review right off the bat yeah I think it’s a great idea to think about targeting it to the right level there are many beginner grants the private agencies like the American Heart Association or the NIH have startup grants that are small and in scope and some you can take a small idea out of your main proposal and and frame it as perhaps a pilot study or a beginning investigator award you can also take a grant that wasn’t funded by one agency and submit it in response to an RFA or

or to a another agency that might have interest in your application so a young investigator just starting out shouldn’t be discouraged by an initial negative result no I would if we were discouraged by negative results they the senior investigators wouldn’t be here either okay if you had to give one piece of advice to a new young investigator what would that be well I think the key thing is to get organized putting together a grant is a major orchestration and it involves a lot of work besides the sitting down in writing and you have to have the big-picture view keep your deadlines in mind and stay organized thank you very much dr Caggiano that’s my pleasure good day my name is Gayle briber I’m an epidemiologist and health services researcher at the University of Washington and the director of the VA summer epidemiology program that’s occurring here at the University of Washington I am happy today to introduce to you some of the participants of the VA epidemiology session and share with you some important characteristics of the VA in general I’ve asked dr. Gordon Starkey baum to join me today dr. Starr Kibum is the director of healthcare at the VA Puget Sound health care system he’s a Rheumatologist he’s a professor of medicine at the University of Washington and he is also associate dean of medicine at the University of Washington dr. Starr Kibum you are affiliated with a very interesting healthcare organization would you tell us a little bit about the VA health care delivery system thank you for inviting me Gayle VA healthcare is is a nationwide and complex undertaking in the United States today there are nearly 25 million veterans and in that group a subset seek care at VA hospitals and facilities all across the country we last year saw nearly four million patients both outpatient and in patients those patients saw providers and and visited nearly 42 million times across the country in facilities at 163 different medical hospitals as well as over 800 clinics throughout the country and the budget for the healthcare part of VA is almost 22 billion dollars there’s 220,000 employees so we’re a very complex large organization providing care to these veteran patients one of the features I think that makes it so exciting and one of the reasons I came to this class is VA is trying very hard to systematize the care of patience and the population of patients that it cares for we we obviously take care of a variety of patients but we also have a very robust data system and we attempt to use that information to provide us information on how well we’re doing and where we can improve care for these patients how is the healthcare system organized in the VA right now there are 22 what are called networks or veterans integrated service networks or visan’s which sounds curiously close divisions and each one of those then has its own governance and facilities for example in our network in the northwest which covers a very large part of the landmass because we also include Alaska we have Alaska Washington Oregon and Idaho we have eight facilities and about 12 I think outpatient clinics and we try to integrate our care so that there is coordination and planning throughout this entire area and that provides a major benefit to veterans who travel in and out in and among those facilities and it’s particularly set up because the tertiary care needs our met primarily in the large urban centers both Seattle and Portland VA medical centers and so a patient for example at the Spokane VA Medical Center who needs cardiac surgery would be offered that surgery at the VA Puget Sound in Seattle in addition are there other don’t a lot of your patients not only benefit from inpatient and outpatient care but also nursing homes home health things like that absolutely

and was we really do have a very large encompassing medical care model from outpatient clinics to inpatients to nursing homes to domiciliaries so a patient who is in with VA cannon in a sense have expected expected to have coordination of his entire spectrum of health care needs from medications to hospitalization to outpatient services radiation surgery and if necessary in some cases nursing home now given that there are 25 million veterans and only 4 million are receiving care in the VA why is it that such a small proportion actually are cared for by the VA well that’s a good question but resources are the short answer we have with 22 billion dollars in the budget we consider ourselves stretched in order to provide good care to those 4 million and you should realize that we are focusing on a sub population of those patients of those veterans so that we’re looking both at elements of service connection as well as their monetary needs and Congress has in a sense made this through a series of priorities and those are based on both consideration of service connection as well as income status and I think what we do focus best on and most on our lower income and patients with disabilities and with special needs and so that is something that has a much higher percentage of veteran patients captured under our health care system so how do you identify and best manage those high-risk patients in the VA system well that is of course a major challenge for us because we do want to husband our our resources properly we want to provide adequate and optimum care and so we we need to pay attention to patients who have special needs and so on I would say that you know we’re still learning about how best to do this we use data we have a computerized system in the VA that captures a number of parameters we can use that information to approach sort of the high-risk patient by that we might mean a patient who has maneuv in the future special needs and expensive needs and so how can we manage for example a patient who may have be heading towards a stroke how can we anticipate that and put things into place to prevent them from having a stroke so that they don’t become such an expensive health burden for our system but one of those strategies be clinical guidelines I think so in other words clinical guidelines reflect a compilation and recommendation by expert bodies of individuals not simply in the VA but some are modified in the VA to say what makes best sense to provide care for these patients and and based on evidence and so on and so using clinical practice guidelines is is a tool and a an approach that we think would be useful to to manage patients now patients with many chronic disease present management challenges how are you working with patients to address some of their educational needs and management challenges I think that’s an excellent question we we know now that simply providing the doctors and the staff with a clinical practice guideline and saying follow this guideline and take care of your patients in in the clinic setting where you have 20 minutes and there are multiple problems to be addressed is a major challenge for us and so I think that we’re coming to understand that we really do need the cooperation and the involvement of patients after all it’s their medical problem and they’re the ones that should benefit from this I think this concept and approaches is been put out by the Institute of Medicine in their crossing the quality chasm report which very much endorses a patient-centered approach to care in in this 21st century and so the challenge in VA is is how do we reach patience how do we educate them how do we approach this issue have because these are complex diseases and there needs to be a great deal of thought and attention put into that I know that the VA has taken an initiative and they’re in looking at patient satisfaction what kind of feedback are you getting well

when we ask questions of our patients by and large they like the care that they get n VA in other words our satisfaction scores in many areas in terms of courtesy attention to their needs and so on are very good they don’t like to wait and so access is often a problem both in the clinic or at the window and the pharmacy or for seeing a specialist or things of that sort and so we’re aware of some of those satisfaction issues and are trying to address those congratulations on the good work that you’re doing and thank you for being with us today dr. Starkey Bob thank you