AFIB Stroke Risk

1 AFIB Stroke Risk Phil Lucht Rimrock Digital Technology, Salt Lake City, Utah 84103 last update: Feb 16, 2015 This is the first of two documents on AFIB risk -- the second concerns bleeding risk. Both documents are located here: http://user.xmission.com/~rimrock/ . The author is not a medical doctor and the unreviewed data and conclusions presented below are just for readers curious about the origins of AFIB risk numbers. If you have AFIB, go see your doctor! Overview .................................................................................................................................................. 2 1. What is a stroke?........................................................................................................... ...................... 2 2. Computation of stroke rate for non-AFIB patients in the AFIB-age population.......................... 5 3. Benefits of Anticoagulation for AF IB patients; Benefits of Ablation............................................ 7 4. What is the CHADS 2 scoring system?............................................................................................. 13 5. What is the CHA 2DS2-VASc scoring system? ................................................................................ 17 6. The QStroke Calculator ...................................................................................................... ............. 23 7. Accuracy and Ambiguity...................................................................................................... ............ 24 8. Long term anticoagulation a fter successful cath eter ablation? .................................................... 28 References.............................................................................................................................................. 29 Abbreviations : AFIB = atrial fibrillation (and flutter) OAC = oral anticoagulant NOAC = "new" oral anticoagulants like apixaban AHA = American Heart Association CI = confidence interval 2 Overview There are many web-accessible articles describing atrial fibrillation and atrial flutter which we group together under the moniker AFIB. In the two documents AFIB Stroke Risk and AFIB Bleed Risk we explore the risks associated with having AFIB and th e benefits of taking oral anticoagulants ("blood thinners", such as warfarin or newer drugs) to reduce stroke risk. The role of catheter ablation is mentioned in passing. The main risk associated with AFIB is stroke, as reviewed below. The ma in risk associated with reducing the risk of stroke with oral anticoagulants is internal hemorrhage, either in the brain or elsewhere. In both documents we make use of a large AFIB drug study to assess these risks for the "average" AFIB patient. We then examine various stratification schem es which attempt to show how a specific AFIB patient's risk differs from that of the average AFIB patient due to factors like age, sex and medical history. It happens that our drug study of interest was f unded by the makers of apixaban (Eliquis), and so we use apixaban as a representative "nov el" oral anticoagulant. There is no intention to promote this drug relative to its currently approved competitors dabi gatran (Pradaxa) and rivaroxaban (Xarelto). Summary Section 1 describes the term "stroke" in some detail. Section 2 then computes the annual stroke risk for non-AFIB people who are in the same age range as most AFIB patients. Our conclusion is that these non- AFIB people have an annual stroke risk of about 1% per year. Section 3 then develops a table (3.12) which compares this background stroke risk to the stroke risk encountered by AFIB patients taking various kinds of oral anticoagulants, or taking no anticoagulant. The average AFIB patient who takes no anticoagulant is faced with a 4-fold increased risk of stroke compared to the background population -- on the order of 4-5% per year. That is to say, the chance of having a stroke is on the order of 40%-50% for each decade of having AFIB. This fact should certainly encourage even the most recalcitrant AFIB patient to do something about the problem. Sections 4, 5 and 6 summarize various stratification schemes which allow an AFIB patient to assess his or her personal stroke risk, which may be more or less than that of the average AFIB patient. Section 7 notes that the statistics of AFIB studies are always weak, and conclusions which are smartly presented with 2 or 3 significant figures of precision in fact have less than 1 digit of accuracy. Strata scheme bin results are roughly ±30% accurate. Pr oblems with definitions of terms are mentioned. Section 8 addresses the risk status of an AFIB patient who has had a "successful" catheter ablation. References are provided in the final section, most of which are freely downloadable on the web. 1. What is a stroke? A stroke refers to the inter ruption of blood flow somewh ere in the brain. This can happen in two ways. An embolus is some object traveling through a blood vessel that is not supposed to be there. It could be a chunk of clotted blood or fat or cholesterol or even an air bubble. If this embolus gets wedged and blocks blood flow, one has an embolism . Blockage could also be caused by a locally formed blood clot (thrombus ). In either case, when this blocka ge occurs in the brain one has an ischemic stroke . 3 ischemic: [ < L < Gk iskhaimos ‘stopping blood’, from iskhein ‘keep back’ + haima ‘blood’.] For AFIB, the main issue is an embolus which is a blood clot -- a thrombus that forms in the AFIBing heart and then flows downstream and becomes an embolism (thus, a thromboembolism ). A mini-stroke or TIA (Transient Ischemic Attack) is an isch emic stroke of shor t duration which might do only limited damage but acts as a warning to the patient. For most stroke studies, including the Apixaban Study discussed below, a TIA is counted as a stroke only if it lasts more than 24 hours. A hemorrhagic stroke is due to a blood vessel that bursts in the brain, allowing blood to leak out and not arrive at its proper destination. This happens either due to vessel ma lformation, or due to a balloon-like vessel bulge ( aneurysm ) which ruptures. hemorrhagic: [ < L < Gk haimorrhagia : 'bleeding violently,' from haima 'blood' + rhage 'a breaking'] Possible causes include high blood pressure, weak vessels, and possibly undetected emboli which "transform", see below. A small such event might quick ly clot and be OK, but anticoagulants prolong the bleeding and thus permit more blood deprivation. Both kinds of strokes can result in an infarct (localized dead tissue) leadi ng to a "deficit" in the patient and possible death. About 87% of strokes are ischemic and 13% are hemorrhagic. (1.1) Here are some AHA stroke pictures: (1.2) http://www.strokeassociation.org/ST ROKEORG/ AboutStroke/TypesofStroke/Typ es-of Stroke_UCM_308531_SubHomePage.jsp A systemic embolism as used below is a embolism occurring some where in the body other than the brain. An example is a pulmonary embolism in the lungs. 4 Sometimes an ischemic stroke transforms (converts) into a hemorrhagic stroke in the sense that that there is bleeding secondary to the causative embolism. It is possible that the original embolism might dissolve before being detected, causing the ischemic str oke to appear as a hemorrhagic stroke. Such hemorrhagic transformations (HT) are poorly documented in IDC-9 ( ≥ 1978) and IDC-10 ( ≥ 1994) hospital codes. An HT event might be coded as an ischemic stroke, as a bleeding event, or both. This fact creates some ambiguity in retrospective studies based on hospital c odings. In the Apixaban Study discussed below, for warfarin patients 20 of 155 ischemic strokes transf ormed (12.9%) while for the apixaban patients 12 of 149 transformed (8.1%). HT therefore provides a pathway for AFIB embolisms, detected or not, to increase AFIB patient bleeding risk, in addition to the direct bleeding risk due to oral anticoagulants. Two varieties of hemorrhagic strokes are indicated in this graphic. Note that both occur on the brain side of the web-like arachnoid membrane. Later we shall discuss bleeds which occur outside this membrane. (1.3) http://www.merckmanuals.com/home/brain_spina l_cord_and_nerve_disorders/stroke_cva/overview_of_hemo rrhagic_stroke.html#v739658 A possible cause of such a stroke is "banging the head ", usually in an unexpected fall. Such traumatic events are usually excluded from stroke and bleed stud y databases, but are counted in strata schemes (see below) which add scoring points for a previous stroke or bleed. Similarly, some TIA's are often excluded from stroke databases, but are counted as a previous stroke in these strata schemes. 5 2. Computation of stroke rate for non-AFIB patients in the AFIB-age population As a baseline, we shall attempt to calculate the average stroke rate for non-AFIB people in a special population that is age-matched to the age distributio n of AFIB patients. Then later when we compute AFIB stroke rates, we can compare these rates to the aged-matched population who have no AFIB, and this then tells us the added stroke danger that is caused by (associated with) AFIB. What is the age distribution of AFIB patients? We assume the AFIB age distribution does not change radically over time, so we can start with this 1995 chart of US AFIB prevalence taken from the 2006 AHA Guidelines (p 11), to which we have a dded some approximate counts in red : (2.1) Our first task is to re-bin the data into bins that are 10 years wide. This gives, (2.2) The "fraction" column shows the percentage in each age bin, and the total number of AFIB patients was 2.2 million in 2006 ( "America"; 4.5 million in the EU). It seems clear that AFIB is primarily a disease 6 of "older people", but not entirely so. Due to the strong peaks in the older age bins, study results are moderately influenced by the averag e age of the participants in the study. This is because older people tend to have more other medical problems ("comorbidity") besides AFIB. One wonders how this distribution will change in the coming decades. We can estimate the average age of the above distribution in this manner: (30*10 + 40*60 + 50*160 + 60*260 + 70*510 + 80*840 + 90*370)/2210 = 73.53. The Apixaban Study has a median age of 70, close to our average number. To obtain stroke rates, we use a certain 2011 AHA update to a 1998 Cincinnati/Kentucky Stroke Study (see Refs). Perhaps the numbers are not exact for th e entire US population, but they are good enough for our rough calculation. ( 2 . 3 ) It is a sad fact that stroke rates are much higher for blacks than whites, and this causes the population - adjusted US stroke rate distribution to be geographically larger in the southeast, as shown by the inset. To simplify our quick analysis, and since blacks are perh aps only 13% of the population, and since the chart states that much of the black data is unreliable, we shall just use the white component of the data. Notice that the age bins match those of our AFIB distributio n in (2.2). Using the numbers in the above graphic, we now compute the expected annual rate of a firs t stroke for the average person in the AFIB-age population (but not having AFIB). It is true that a tiny fraction have AFIB, but we ignore that fact. 7 (2.4) We enter the white men and white women rates from the chart, average them in the W column, then weight that result by the AFIB population fraction. The conclusion is that about 11.54 per 1000 of the AFIB-age population have a first-ever stroke each ye ar. Therefore (AFIB prevalence is ~ 2/300 < 1% so can be neglected here), Average stroke rate for non-AFIB patients in the AFIB-age population = 1.15 %/year (2.5) 3. Benefits of Anticoagulation for AFIB patients; Benefits of Ablation Warfarin properly managed reduces the risk of a stroke for people having either intermittent (paroxysmal) or continuous (persistent) AFIB. Most AFIB-related strokes are thought to arise from blood clots forming in the left atrial appendage during AFIB which then fl ow through the left ventricle to the brain. Below is data from the 2014 AHA AFIB Guidelines (p 31) on the stroke reduction benefit of taking warfarin versus taking nothing at all (control group, placebo). Since some in the control group were taking aspirin, the results are more dramatically in favor of warfarin than the data below indicate: (3.1) 8 If we give less weight to the CAFA result which has a huge error bar, an eyeball average of the results suggests that warfarin provides a 64% reduction in the rate of strokes compared to taking nothing, so one writes Risk Reduction = RR = 0.64. If a rate r 1 is reduced by some RR to give r 2 then r2 = r1 (1-RR). The Risk Reduction Factor would then be RRF = r 1/r2 = 1/(1-RR). For example, if RR = 80%, then RRF = 5, and one would say the rate was reduced by a fact or of 5. For the above, RR = 0.64 so RRF = 1/.36 ≈ 2.78 and so warfarin cuts the stroke rate by a factor of about 3! This figure is so dramatic that no new similar studies will ever be done on ethical grounds (last was 1993). For example, the BAATAF Boston study was terminated early when the large ~ 300% benefit of warfarin was detected. Support for a 64% stroke rate reduction for warfarin (and 22% for aspirin) comes from Hart et al. (2007) from which the above graphic was taken (without our red line). Hart did a meta-analysis of 29 trials involving a total of 28,140 AFIB patients. Th ese Hart reduction rates 64% and 22% are often quoted in the literature. Recently the 22% rate reduction for aspirin has been challenge d, especially for older patients, as coming from weaker studies which overestim ate its stroke reduction benefit, but we shall assume it applies below. So, stroke reduction rate for warfarin 64% stroke reduction rate for aspirin 22% (3.2) The Apixaban Study The large Apixaban Study (2011, N=18,201) had tw o cohorts of about ~90 00 patients each, one cohort taking apixaban (a new oral anticoagulant, brand name Eliquis) and the other taking the traditional warfarin (Coumadin). This study, known informally as ARISTOTLE, was funded by the apixaban makers Pfizer and Bristol-Myers Squibb. Here are the str oke rates found in this study for the two cohorts, ( 3 . 3 ) We interpret this data to read: Apixaban Warfarin ischemic stroke 0.97 %/yr 80% 75% 1.05 %/yr 69% 65% hemorrhagic stroke 0.24 %/yr 20% 18% 0.47 %/yr 31% 29% 1.21 %/yr 100% 1.52 %/yr 100% systemic embolism 0.09 %/yr 7% 0.10 %/yr 6% total stroke rate 1.30 %/yr 100% 1.62 %/yr 100% (3.4) 9 Presumably these numbers are for strokes which began as indicated. Some of the ischemic strokes "transformed" into hemorrhagic ones due to bleeding secondary to the embolism blockage. Notice that the percentage of strokes that are hemorrhagic for these anticoagulated patients (20% and 31%) are much higher than for the baseline population taking no anticoagulation (13%). The systemic embolism rate contributes a small amount to the sums as shown. If we include such embolisms in a broad definition of "stroke", we find that, for the average AFIB patient, r a t i o stroke risk while taking apixaban 1.30 %/year 0.29 stroke risk while taking warfarin 1.62 %/year 0.36 stroke risk while taking aspirin 3.51 %/year // = 4.50*.78 0.78 stroke risk while taking nothing 4.50 %/year // = 1.62/.36 1.00 (3.5) The first two numbers come from the bottom line in (3 .4). The fourth number 4.50 is then obtained from the warfarin number 1.62 using the 64% reduction of st roke rate attributed to warfarin in (3.2), r 1 = r2 /(1-RR) = 1.62/(1-0.64) = 1.62/0.36 = 4.5 . Once this number is known, (3.2) says the aspirin stroke rate is r = (1-.22)4.5 = .78*4.5 = 3.51. The fourth line of (3.5) should be eye-popping to an y AFIB patient. It says that if no OAC is taken, the probability of having a str oke is 4.5% per year or 45% per decade and 90% over 20 years. This is very bad news. Footnote on Apixaban Study Numbers : Here is a version of (3.4) showing event counts. The first two rows' numbers come from page 7 of the Study, while the last two rows' numbers are taken from (3.3) : Apixaban Warfarin ischemic stroke 149 155 uncertain stroke type 14 149+14 = 163 21 155+21 = 176 hemorrhagic strokes 40 78 systemic embolisms 15 1 7 2 1 8 2 7 1 ( 3 . 6 ) For some unspecified reason, the sums here differ slightly from the sums 212 and 265 shown in (3.3). Probably for the same reason, our total rates 1. 30 and 1.62 shown in (3.4) differ from 1.27 and 1.60 shown in the first line of (3.3). This could be relate d to patients who left the study prematurely. We shall base our work on the 1.30 and 1.62 numbers for total stroke rate shown in (3.4). Now, using the result (2.5) that the non-AFIB stroke ra te as 1.15 %/yr and the (3.5) rate of 4.50 %/yr, we reach this conclusion: stroke rate for the average AFIB patient (no-OAC) stroke rate for the average non-AFIB but age-matched person = 4.50 1.15 = 3.91 (3.7) 10 It would be nice to have some external support for th is AFIB/noAFIB stroke ratio of 3.91. Our derivation of the ratio is specific to the AFIB age distri bution (2.1) and the stroke distribution (2.3). The Framingham Study: just what is the AFIB/noAFIB Stroke Ratio? Since ethical modern trials only include coagulated AF IB patients (due to the 4.5 %/yr number above), we have to look back at older work, and a key piece of older work is known as the Framingham Study (Framingham is a town west of Boston). In this im pressive study 5,209 AFIB patients were followed up every two years from 1948-1982 (34 years!). The goal was to study the rates of incidence and prevalence of AFIB (versus age, for example) and to learn more about the association of atrial fibrillation with other medical conditions, and with stroke. Various papers have been written analyzing this study, and we show three of them in our References. In a 1987 paper, Wolf, Abbott and Kannel say "In recent years, chronic atrial fibrillation in the absence of rheumatic valvular heart disease has been found to be associated with more than a fivefold increased incidence of stroke, even when age and hypertensive st atus are taken into account." Thus enters the notion of "5-fold" for the AFIB/noAFIB stroke ratio, and this ratio appears in the 2014 AHA Guidelines (p 27 4.1.1). The 1987 paper indicates that anticoagulants like warfarin were "seldom used". In their Table 3, the authors present the following values of our ratio of interest for four different age bins: (3.8) Notice the fairly large error bars on these numbers (see Section 7 for general comments on error bars). The paper says that the ratio is 4 for "lone AFIB" patients -- those having no heart disease other than AFIB. In a later 1991 paper, the same authors present a si milar chart showing the age-binned ratios, but the four numbers are lowered slightly to 4.0, 2.6, 3.3 and 4.5 which appear in the bottom row below, (3.9) If we average the four numbers, the result is (4.0+2.6 +3.3+4.5)/4 = 3.6. If we weight the numbers by the AFIB prevalence counts in Table 1 of the 1991 paper, th e mean ratio is 3.36 as shown on the left below. But in this same 1991 paper, the summary describes th e ratio as "near fivefold", and the paper's Figure 1 suggests the ratio is 4.8. Alternatively, we can we ight the ratios of the 1991 paper by our 1995 population distribution (2.1) and this gives 3.62 as shown on the right below : 11 (3.10) Here we assume that half the patients age > 85 are in the range 85-89, but the result is not very sensitive to this assumption. In a still later 1998 paper of Kannel, Wolf, Benjamin and Levy the ratio is stated as "4-5-fold" in the summary, but later in the text they say "Epidemiologi c and clinical studies have generally indicated that AF constitutes a major independent risk factor for str oke, with a 3–5-fold increased risk after adjusting for other risk factors (Table V)", and Table V restates the data of (3.9) above. In fact, if we weight the four ratios using the population distribution stated in Fig 1 of the 1998 paper, we find an average ratio of 3.90: ( 3 . 1 1 ) Based on this tour of the various Framingham analyses, we feel that for the popu lation distribution (2.1), the average ratio of 3.91 shown in (3.7) is "reasona ble". Based on the comments of Section 7 below, a true statement would probably be that the ratio is 4 ± 1, but we shall continue with our 3.91 number. Conclusion on average stroke rates for various OAC Having validated the 3.91 ratio as being "reasonable", we go ahead and extend our table (3.5) by adding a first row for non-AFIB but age-matched people, us ing the 1.15 %/yr non-AFIB rate from (2.5), (3.12) 12 The rightmost 5 columns show rate ratios normalized five different ways to make it easy to read off conclusions from the table. The ratio 3.91 of (3.7) is then visible at the bottom of the first ratio column. Here are a few conclusions from this table: • Warfarin provides a factor of 2.78 reduction in total stroke risk (isch+hem+emb) compared to no anticoagulation, and reduces the AFIB patient's stroke risk to a point only 41% higher than that of an aged-matched non-AFIB person. (3.13) • Apixaban provides a factor of 3.46 reduction in total stroke risk (isch+hem+e mb) compared to no anticoagulation, and reduces the AFIB patient's stroke risk to a point only 13% higher than that of an aged-matched non-AFIB person. (3.14) The ratios indicate that apixaban decreases stroke risk by 20% compared to warfarin, while taking nothing increases stroke risk 278% compared to warfarin. Th is then explains why it is that AFIB people historically have taken warfarin and why they might be interested in the "novel" anticoagulation drugs (NOACs) such as apixaban (Eliquis). (3.15) The rates shown are for an average patient. If patients are stratified according to so me risk factor(s), such as in CHADS 2 scoring (see below), perhaps the ratios above stay constant. Then for example if one's stroke risk due to having some set of risk factors (hyp ertension, previous stroke, etc.) is twice the average, the stroke rates in (3.12) would all be multiplied by 2. The good news then is that anticoagulation brings the stroke risk of the AFIB patient down to a level not too much larger than that of the same-age non-AFIB population. The anticoagulated AFIB patient stroke risk is not 4-5 times that of the non-AFIB patient risk. How does this all relate to catheter ablation? Assuming that ablation completely restores a patient to the non-AFIB baseline (which is questioned, see Section 8 below), the post-ablation patient no longer taking anticoagulants has a stroke risk 29% (12%) lower than he or she had while taking warfarin (apixaban) before ablation (this from (3.12) ratios). This is not a particularly dramatic rate decrease, but th e post-ablation patient no lo nger taking anticoagulation has a dramatically reduced bleeding risk, as discussed in our second document. Unfortunately, recent Guidelines recommend that anticoagulation be continue d indefinitely after catheter ablation for people with CHA 2DS2-VASc score ≥ 2 (discussed below), so for those people the bleeding rate is not reduced, and the stroke rate is only moderately reduced by ablation. But even for those patients, assuming they have active AFIB, there are still si gnificant benefits for ablation: • removal of the discomfort of AFIB and of re duced heart output making patients weak and tired • lowered risks of dementia and heart problems (a nd mortality) resulting from the heart's reduced output and general malfunctioning during AFIB. If a patient is unable or unwilling to take anticoagula tion drugs, we can add this third ablation benefit : • lowered stroke rate by a very significant factor of about 4 (3.16) 13 4. What is the CHADS 2 scoring system? This stratification scheme allows one to compute the risk multiplier like the factor of 2 mentioned above. The original CHADS 2 system was concocted in 2001 by Gage et al . Each letter of the mnemonic CHADS 2 stands for a factor which increases an AFIB patient's risk : (read the capital letters down) // 2014 AHA Guidelines (4.1) The subscript 2 on S 2 means that a previous Stroke gets 2 points in the scoring, while other factors get 1 point. For example, if you have hypertension, are over age 75 and had a TIA, your CHADS 2 score is 4 out of 6. This study dealt with ischemic strokes in a patie nt set called NRAF (National Registry of Atrial Fibrillation) that was relatively old and sick and c ould thus give good statistics for the various risk factors. There were 1,733 AFIB patients, none of whom took warfarin, and 31% of whom took aspirin. Average age was 81. For those 1,204 in the set taking no OAC the average CHADS 2 score was 2.1, while for the 529 who took aspirin, average score was 2.3. The Gage paper calculations of annual stroke risk (%) versus CHADS 2 score are shown in the rightmost column below: (4.2) The authors' footnote † presumably implies the rates were adjusted to remove the aspirin effect, so it is as if all 1733 patients were taking nothing. 14 Roughly, stroke risk increases by a factor of 1.5 for each unit increase in the score. In fact, a nearly perfect fit to the 7 rates shown in the right column in (4.2) is obtained from annual no-OAC stroke risk rate = 1.9 * (1.457)score or rate(s) = 1.9 * (1.457)s . (4.3) In the graphic below, the fit (4.3) is shown as a red curve, while the black squares show the rate values from the right column of (4.2), and one can see that th e red curve fits the data very closely. The dashed curves show the error bands also from (4.2). (4.4) Superposed on this plot is a blue histogram showing the relative number of patients in the study for each score bin, taken from the second column of (4.2). If the red rate curve displays the function rate(s) where s is the score, and if the histogram bar heights are fractions f i such that Σifi= 1, then we claim that [ here "average" means for the average person in the study ], average rate ≡ <rate> = Σs fs r a t e ( s ) average score ≡ <s> = Σ s fs s . ( 4 . 5 ) Since the red rate curve is non-linear , the value of the rate function evaluated at the average score does not equal the average rate! That is to say, rate(<s>) ≠ <rate>. If the rate happens to be a linear function rate(s) = ms + b, then these two quantities are in fact equal, since rate(<s>) = m <s> + b = m Σs fs s + b <rate> = Σ s fs rate(s) = Σsfs [ms+b] = m Σs fs s + (Σsfs)b = m Σ s fs s + b . (4.6) 15 In our stroke and bleed documents we shall be seeing several of these stratification schemes, and the rate curves are never linear, so it is important to unde rstand the fact that in general rate(<s>) ≠ <rate>. Thus, if one wants to know <s> and <rate>, one has to compute each separately. We now do exactly this using the data shown in (4.2) and the expressions shown in (4.5), (4.7) The columns are s = score, r = rate, n = number of study patients with score s, f = fraction f s of patients at that score (obtained from the n column and its sum 1733), and the last two columns are f s*s and f s* rate(s) which appear in (4.5). Adding these columns then does Σs and we find the average score and the average rate to be 2.17 and 4.83. As a check on our non-equality of the two quantities, we find that rate(<s>) = rate(2.17) = 1.9 * (1.457) 2.17 = 4.29 ≠ <rate> = 4.83 (4.8) With this long-winded introduction (which we shall not repeat for later strata schemes), we now display the following stroke risk table based on the NRAF study, but where the rates are scaled by an overall scale factor sf = 0.93: ( 4 . 9 ) We now explain this table in some detail. The scale fact or 0.93 (next paragraph) is in the leftmost column, followed by the rates "r" taken from the right column of (4.2). The bottom entry of this column is the average rate 4.83 taken from (4.7). The r column rat es are then scaled down by the scale factor to give rates in column sf*r. These scaled-down rates are th en copied over to the no-OAC column under the number 1.00. The CHADS 2 score is shown in the CHADS2 column . The remaining 4 columns are then obtained from the numbers in the 1.00 column by scali ng with the fractions shown across the top (.78, .36 and so on). These fractions come from the last column of (3.12), so we are assuming the relative stroke rates among the different OAC's are the same at each CHADS 2 score. 16 Now why this 0.93 overall scale factor? This causes ta ble (4.9) to be calibrated to the Apixaban Study population. Recall from that Study as shown in (3.4) that the average stroke rates for apixaban and warfarin patients were 1.30 %/yr and 1.62 %/yr. The factor 0.93 causes the bottom numbers in the apixaban and warfarin columns in (4.9) to match th ese values. Once these match, all the average rates match those in (3.12). [Although the absolute CHADS 2 rates are for ischemic stroke, we calibrate (4.9) to match the total stroke numbers of (3.4)] The left no-OAC column in (4.9) shows stroke rates for AFIB patients at each CHADS 2 score level who are not taking any anticoagulant. The rightmost no-OAC column shows rates for the age-matched non-AFIB population. Here then is a plot of the data in table (4.9) : Annual Total Stroke Rate CHADS 2 score → (4.10) The benefits of taking oral anticoagulants (OAC) for AFIB especially at higher CHADS 2 scores could not be more dramatic. It is helpful to multiply the risk by a factor of 10 to get risk per decade of having AFIB. For example, a person with score 4 taking no OAC has an 79% chance of having a stroke over 10 years. Example: A relatively healthy AFIB patient has a CHADS 2 score of 1: • taking no OAC, this patient has a 26 %/decade risk of stroke • taking warfarin, the risk drops to 9 %/decade • taking apixaban, the risk drops to 8 %/decade • a corresponding no-AFIB patient has a risk of 7 %/decade (4.11) Comment : The scale factor 0.93 required to calibrate th e NRAF data to the Apixaban Study data might be attributed to the fact that the average age of the NRAF group was 81, while the Apixaban Study had an median age of 70, so we are not surprised to see that the NRAF results have to be scaled down a bit. 17 5. What is the CHA 2DS2-VASc scoring system? Lip et al. (Lip is from Birmingham, UK) came up w ith a more refined version of the CHADS 2 strata scheme which better accounts for the effect of Age a nd the Sex of the patient (older and/or female increase stroke rate) and adds a Vascular risk factor. This new schema is usually pronounced "chads-two- vask" and here is the scoring plan (from Lip et al. 2010 "Refining Clinical Risk...") : (5.1) In this refinement, previous stroke S still gives 2 scoring points, but now age ≥ 75 also gives 2 scoring points, hence the two "2" subscripts A 2 and S2. Presumably the "c" suffix on the last letter was installed to distinguish that S from the S for stroke. There are now two age measures instead of one, called A and A2, so age can add 0, 1 or 2 points to a score. Whereas the CHADS 2 score ranged 0 to 6, the CHA 2DS2- VASc score ranges 0 to 9, and women cannot have a 0 score even if under age 65. Lip et al. (2010, "Identifying patients...") then present this table of stroke rate versus CHA 2DS2- VASc score for patients on warfarin in an amalgamated study group of 7,329 AFIB patients: [ PY = patient-years. TE = thromboembolic ] 18 (5.2) • Unlike the CHADS 2 table (4.2) (which was for patients taki ng no anticoagulation), the 2nd last column stroke rates of (5.2) are for patients taking warfa rin. Using the warfarin RR = 64% reduction idea noted earlier above (3.2), with corresponding RRF = 1/(1-RR) = 1/.36 = 2.78, Lip gives rates in the last column as 2.78 times the rates in the second last column. • Unlike the CHADS 2 score, a patient with score 0 (a male AFIB patient under age 65 with no risk factors) gets a stroke risk of 0% and can forego anticoagulation. • Unlike the CHADS 2 score, the stroke risk percentages are not monotonically increasing with score, which seems a bit strange -- a fluke of low statistics we presume. Consider these two plots, (5.3) 19 The black points are the actual Lip rates from the 2nd last (warfarin) column of (5.2). The red curve is the following simple fit to the data : rate = 3 tan(score/8) // red curve, tan argument is radians not degrees (5.4) The fit is "reasonable" except for score = 8 which seem s to be an outlier since rate goes down there as the score goes up. Certainly the red curve is "legal" in terms of the upper and lower 95% confidence interval boundaries which are included is this plot as dashed lines (we lowered the top right x from 27 by 10 to make things more visible) (5.5) The above graphs come from the following spreadsheet : (5.6) The (d-f)/d column shows that the percent alteration of the fit from the original data is within 20% except for the outliner point with score 8. The data in the l ast two columns are 2.78 times the data in the warfarin columns, as noted above. We shall use the more "logical " monotonic tangent fit data in our work below. 20 Footnote. The Lip et al. cohort of 7,329 patients was obtained from two studies called SPORTIF III and V. The people in these trials were taking either warfarin or ximelagatran (Exanta, Exarta), a trial anticoagulant. When the two SPORTIF studies were co mbined, warfarin and ximelagatran had about the same risk for preventing stroke, so we can just pret end that all 7,329 patients were taking warfarin. Thus, the 2nd last column stroke rate percentages shown above in (5.2) are (in effect) for patients taking warfarin. It happens that ximelagatran was withdraw n from trials in 2006 due to liver enzyme issues in 5% of patients. This reminds one that a modest dela y in starting even an approved new replacement drug might be something to consider. Using the same method leading to (4.7), we now compute the average CHA 2DS2-VASc score and rate: (5.7) Then using the average rate 1.62 we construct a stroke risk table in exact analogy with the CHADS 2 table (4.9), (5.8) The rates in the r column are the ad justed fit rates shown in (5.6). As we shall see, this time no scale factor is required, so we set sf = 1.00 and then the rates from the r column just copy to the sf*r column. They are copied once again to the warfarin column (since the data in this study are for warfarin patients, whereas the CHADS 2 study was for no-OAC patients). The rema ining columns are then filled out using the multipliers shown at the top (2. 78, 2.17, etc) which come from the middle ratio column of (3.12). 21 In (5.7) the average rate came out 1.62 which happ ens to exactly match the 1.62 average rate of the Apixaban Study as shown in (3.4). For this reason, the bottom row of (5.8) exactly matches the rates shown in (3.12), and that is why no extra scale factor is required. So the Lip data is already "calibrated" to the Apixaban Study. This is largely due to the fact that the Lip amalgamated data set and the Apixaban data sets have the same mean age. Comment : It is of course just a coincidence that the average rate came out 1.62 exactly matching the Apixaban Study number to 3 significan t figures, especially in light of our crude tangent fit adjustment and the usual wide error bands shown in (5.5). Had this exact match not occurred, we would have added a scale factor in the left column of (5.8) to effect the calibration to the Apixaban Study as we did for CHADS 2 in (4.9). Here is a plot of the data in table (5.8) : Annual Total Stroke Rate CHA 2DS2-VASc score → (5.9) The highest (worst) curve is for the AFIB patient taking no anticoagulation. The lowest (best) curve is for age-ma tched people who don't have AFIB. Unlike the CHADS 2 plot (4.10), all five curves meet at the origin since (5.2) indicates zero stroke rate for score 0. Again, the benefit of taking oral anticoagulants (OAC) for AFIB especially at higher CHA 2DS2-VASc scores is dramatic. For example, a person with sco re 6 taking no OAC has a 78% chance of having a stroke over 10 years. 22 Example: A relatively healthy AFIB patient has a CHADS 2-VASc score of 2: • taking no OAC, this patient has a 21 %/decade risk of stroke • taking warfarin, the risk drops to 8 %/decade • taking apixaban, the risk drops to 6 %/decade • a corresponding no-AFIB patient has a risk of 5 %/decade (5.10) The "taking nothing" column of (5.2) and the last column of (4.2) for CHADS 2 appear on page 27 of the 2014 AHA Guidelines, but the Guidelines fail to note th at this data is for no anticoagulation : // no anticoagulation // no anticoagulation (5.11) Our tangent fit CHA 2DS2-VASc no-OAC rates are shown to the right of the Guidelines numbers, taken from the right column of (5.6). Taking a Warfarin Break and Pill-in-a-Pocket ? ( 5 . 1 2 ) For our Example (5.10) -- AFIB patient with a CHA 2DS2-VASc score of 2 -- the annual stroke risk is about 0.8 % if on warfarin, and 2.1% off warfarin. Since these are per-year risk rates, going off warfarin for a few days for dental work or whatever represents only a tiny stroke risk. For example, if one's INR drops to 1.0 for 2 days, the absolute stroke risk for those days would be (2/365)*2.1% = .01% = 1/10,000. For people who rarely go into AFIB (as far as they know), this does suggest the "pill in a pocket" idea for warfarin use. One only takes warfarin on those ra re occasions when AFIB o ccurs. There are doubtless counterarguments to this plan based on the idea that cl ots could in theory form in a few minutes or hours while it might take 2 to 4 days for the warfarin INR 2-3 therapeutic range to be restored. Moreover, the "AFIB begets AFIB" philosophy argues for a quick el ectrocardioversion soon after AFIB occurs, but the 2014 AHA Guidelines (p 52) recommend warfarin be take n "for at least 3 weeks" prior to such a cardioversion. This delay might increase the chances of electrocardioversion not working or increase the patient's paroxysmal recurrence frequency. 23 6. The QStroke Calculator Algorithm s like QStroke are likely to become more prevalent since they are trivial to use online or perhaps with a smartphone app. You quickly enter your particular data and it computes risk based on your data. Such methods can allow for arbitrarily complicat ed models based on "machine learning" (Big Data approach) where one does a logistic regression fit to ma ny factors ("predictors"). QStroke is available as free software under a Gnu public license. Here is a data entry Example ( http://www.qstroke.org/ ) : [ notice the AFIB check box ] ( 6 . 1 ) And here is the result of pushing the button: (6.2) In the two examples (4.11) and (5.10) given earlier (all examples apply to the same patient), being on warfarin with a CHADS 2 score of 1 predicted a stroke rate of 0.9 %/yr, while being on warfarin with a CHA2DS2-VASc score of 2 predicted an annual stroke risk of 0.8 %/yr. Both these numbers are quite close to the QStroke prediction of 0.8 %/yr. It is always encouraging when different methods lead to the same result. 24 7. Accuracy and Ambiguity Accura cy of AFIB Study Results Here we momentarily pull back the curtain and take a brief look at the seamy underside of AFIB trials and studies, which is their accuracy, then we quick ly restore the curtain and look the other way. Most of the studies/trials regarding AFIB stroke (and bleeding) have insufficient statistics to be truly conclusive in terms of the quoted percentage risk rates, as shown for example in Fig (5.5) where the dashed curves show the error band. Un certainty is presented in term of a 95% confidence interval (CI), or perhaps the standard deviation (SD) is given, or error bars are simply displayed as in our Fig (3.1) above, The relative largeness of the error bars is due to the fact that: (1) a study is hard pressed (by cost and time) to get more than a few thousand AFIB patients (2) if these patients are classified into strata scheme bins, the number per bin is smaller still (3) the annual risk rates for stroke or bleed ing are relatively small, on the order of 1-5% To show how these three items conspire to cause trouble, we consider a simple model for the statistics of a trial known as the Bernoulli Trial model. A study of N patients with outcome stroke or no-stroke can be treated like flipping a weighted coin N times where p is the probability of getting a head (a stroke) on one toss (watch patient for a year). After doing the N flips (carrying out one AFIB study), by counting the number of heads H one will find that the fraction H/N is close to p, the mean. The probability distribution for the fraction of heads is the binomial distribution which has mean p and SD = σ mean = p(1-p)/N . Since the binomial distribution is close to the nor mal (gaussian) distribution for large N, the 95% confidence interval is 1.96 σmean in either direction. Thus one would describe the study outcome (p = percentage of patients having strokes in one year) as lying with 95% confidence in this interval, p ± 1.96 σmean = p ± 1.96p(1-p) N = p ( 1 ± 1.96(1-p) Np ) = p ( 1 ± fractional error) fractional error = 1.96(1-p) Np ≈ 1.96 1 Np (if p is small) (7.1) Here we see the penalty inflicted on the frac tional error by small N and/or small p. 25 The following spreadsheet computes fractional erro r as a function of N and p for this model : ( 7 . 2 ) For example, if the mean value of a stroke rate is 4.0% and the patient count associated with that rate is N=1000, then with a 95% CI one would express that rate as( 1 ± .30)*4.0% and one would say the mean value was fractionally accurate to ± 30%. The actual CI range would be [ 0.7, 1.3] 4% = [2.8%, 5.2%]. In actual studies, the Bernoulli model is replaced by a model which gives somewhat smaller error bars. For example, in the censured exponential survival model used in the CHADS 2 study (footnote in (4.2)), one observes the actual times at which strokes occur dur ing the study. This information, ignored in the Bernoulli model, allows for smaller and asymmetric error bars. However, the error bars are in the same general range as the Bernoulli model bars a nd do not do much to reduce uncertainty. The bins in the CHADS 2 and CHA 2DS2-VASc studies have from N ≈ 100 to N ≈ 1700 patients. In our bleed document we deal with similar strata schemes called HEMORR 2HAGES and HAS-BLED which have similar binned patient counts. For each strata sc heme, we consider one particular study (normally carried out by the inventor of the scheme), and here we show the fractional error on the low and high side of the data (felo and fehi) for each of our four strata cases: (7.3) 26 For HAS-BLED the fractional errors are symmetric because we generate them ourselves using the Bernoulli model above, but in the other cases the fracti onal errors are computed from the absolute error bands provided by the study. The large sizes of these fractional errors are discouraging, and are due to the low bin N counts as noted above. If we ignore the HEMORR 2HAGES study errors, we might generously claim that, for the non- extremal scores, the other three strata schemes produce rates with a ballpark ±30% fractional error at 95% CI. Dropping down from two to one standard deviati on, these rates are roughly ±15% accurate at a 68% confidence interval. Thus, the rate numbers appearing for example in tabl e (5.8) presented gloriously with 2 or 3 significant figures really have less than 1 significant digit of accuracy. For example: 0.78 → 0.78 ±.30*0.78 = 0.78 ± 0.23 ≈ 0.8 ± 0.2 95% CI 3.16 → 3.16 ±.30*3.16 = 3.16 ± 0.95 ≈ 3 ± 1 95% CI 6.88 → 6.88 ±.30*6.88 = 6.88 ± 2.06 ≈ 7 ± 2 95% CI In the last case, if the result can be 5,6,7,8 or 9, one cannot claim there is even one digit of precision in the result. That would require the result being 7 ± 0.5 or better. For 68% CI one has 7 ± 1, and still there is not one digit of precision. So we give up on making any fancy claims (such as a "half digit" of precision) and just stick with the ±30% for 95% CI and ±15% fo r 68% CI as our ballpark st rata scheme error ranges. If one wants something more accurate for errors, one can look at (7.3). Nevertheless, we maintain the 3 digit notation to ma ke it easier for the reader to follow the numbers as they get moved around between our various spreadsheet s, and also to reduce any further error arising from computation. Numbers derived for an average patient in an arm of the Apixaban Study with N ≈ 9000 might be accurate to ±10% with 95% CI, just looking at the graph in (7.2). Our rate tables are all based on OAC ratios such as those appearing across the top of (5.8) and of course these ratios also have error bars. For example, compar ing apixaban to warfarin the Apixaban Study gives stroke rates of 1.27 and 1.60 in the top row of (3.3), a nd the ratio of these rates is .79 but the error range is (.66, .95) giving a low and high side fractional error of (16%, 20%) with 95% CI. At least this is better than the ±30 % range for the strata scheme studies . So we might generally claim that any statement regarding ratios of rates for different OAC values mi ght be accurate to about ±20 % fractional error with 95% CI. Ambiguity in AFIB Study Results Everything should have a clean definition but often clean definitions are not provided, are not available, or differ from study to study. For example, what ex actly is a "stroke"? Usually it means a blood flow blockage (ischemic) stroke, but it might include a bleedi ng stroke (hemorrhagic stroke). Normally it is in the brain (cerebral), but it could be elsewhere such as in the lungs (pulmonary embolism). Does a TIA 27 count as a stroke? Do the stroke effects have to last more than 24 hours for the stroke to be counted? Are strokes induced by trauma counted? How does one d eal with ischemic strok es which transform into hemorrhagic ones? Often retrospective studies use historical hospital IDC code data and things are not always clear. If a stroke patient is discharged on warfarin, should one assume he or she was on warfarin during the time period leading up to the stroke? Some other examples: in a CHADS type scoring, does "Hypertension" mean the patient would have high BP without drug control, or does it mean he/she has high BP as a diagnosis for a controlling drug, or does it mean the person actually has high BP? If hypertension means systolic > 160, does that refer to the patient's average value BP over a day, or the awake day, or does it refer to a peak value during a day? In a similar vein, exactly what is alcohol abuse in the HEMORR 2HAGES and HAS-BLED scorings which appear in our bleed document? In our AFIB bleed document there are similar ambi guities for the definition of a "major bleed". The Apixaban Study felt compelled to present results by th ree different major bleed standards, and each of the two strata schemes (HEMORR 2HAGES and HAS-BLED) has its own s lightly different definition. Probably events counted by all these definitions are very similar. 28 8. Long term anticoagulation after successful catheter ablation? This see ms to be a delicate subject right now. The 2014 AHA Guidelines are silent on the topic. However, here is what the 2012 ESC Guidelines have to say regarding peri- and pos t-ablation anticoagulation (p 2741): (8.1) The problem is a lack of data (so far) on AFIB recurrences after "successful" ablation, and on post- ablation stroke rates for those on and off long term OAC (oral anticoagulation). Doctors have seen post- ablation people go off OAC and die, for example. It is possible that "silent AFIB" will start up undetected by the patient, or regular symptomatic AFIB might start up. Things are not well understood, and some doctors don't want to take the risk of patients going off OAC for CHA 2DS2-VASc scores ≥ 2. One approach would be to wait a year, then do a 30 day Holte r test, then maybe go off if it is clean. Here is another comment on the subject (Levitt, 2009) in which an elevated CHADS 2 risk means a score ≥ 2 : ( 8 . 2 ) The ESC Guideline is more stringent, requiring long-term post-ablation OAC for a CHA 2DS2-VASc score ≥ 2 . Hopefully more data will be collected on this subject. 29 References Current publi c-free-access web links are given for most papers. Where possible, Digital Object Identifier links of the form doi = prefix/suffix are also given for use as http://doi.org/prefix/suffix. References are listed in the order first encountered in the our document. [ all links verified 13 Feb 2015 ] _____________________________________________________________________________________ • 2006 AHA AFIB Guidelines ( for age chart (2.1) ) Fuster V et al. ACC/AHA/ESC 2006 Guidelines for the Manageme nt of Patients With Atrial Fibrillation, Circulation 2006;114:e257-e354 [doi: 10.1161/CIRCULATIONAHA.106.177292] http://circ.ahajournals.org/content/114/7/e257.full.pdf+html _____________________________________________________________________________________ • Cincinnati Stroke Study Broderick J et al. The Greater Cincinnati/Northern Kentucky Stroke Study, Stroke 1998;29:415-421. http://stroke.ahajournals.org/c ontent/29/ 2/415.full.pdf+html However, our Fig (2.3) appears as a black and white Chart 4-4 in the following 2011 update, Véronique LR et al. Heart Disease and Stroke Statistics—2011 Update : A Report From the American Heart Association, Circulation February 1, 2011 vol. 123 no. 4 e18-e209. [doi: 10.1161/CIR.0b013e318200970] http://circ.ahajournals.org/conten t/123/4/ e18/F46.expansion.html The actual color figure we used came from here: https://www.heart.org/idc/groups/heart-public/@wcm/ @sop/@smd/documents/downloadable/ucm_449858.pdf . _____________________________________________________________________________________ • 2014 AHA AFIB Guidelines January CT et al . 2014 AHA/ACC/HRS Guideline for the Management of Patients With Atrial Fibrillation: A Report of the American College of Ca rdiology/American Heart Association Task Force on Practice Guidelines and the Heart Rhythm Society, J Am Coll Cardiol. 2014;64(21):e1-e76. [doi:10.1016/j.jacc.2014.03.022] http://content.onlinejacc.org/ article.aspx?articleid=1854231 . _____________________________________________________________________________________ 30 _____________________________________________________________________________________ • OAC Reduces AFIB Stroke Rate (Hart et al.) Hart RG et al . Meta-analysis: Antithrombotic Therapy to Prevent Stroke in Patients Who Have Nonvalvular Atrial Fibrillation, Ann Intern Med. 2007 Jun 19;146(12):857-67. _____________________________________________________________________________________ • Apixaban Study (ARISTOTLE) Granger CB et al. Apixaban versus Warfarin in Patients with Atrial Fibrillation, N Engl J Med 2011; 365:981-92. [doi: 10.1056/NEJMoa1107039 ] http://www.nejm.org/doi/full/10.1056/NEJMoa1107039 [ A Supplement containing contributor names and event definitions appears also at the above link.] _____________________________________________________________________________________ • Framingham Study Wolf PA et al. Atrial Fibrillation: A Major Contributor to Stroke in the Elderly: The Framingham Study, Arch Intern Med 1987 Sept;147:1561-1564 [ doi:10.1001/archinte.1987.00370090041008 ] Wolf PA et al. Atrial fibrillation as an independent risk factor for stroke: the Framingham Study, Stroke 1991;22:983-988 [doi: 10.1161/01.STR.22.8.983] http://stroke.ahajournals.org/content/22/ 8/983.short?r ss=1&ssource=mfc Kannel WB et al. Prevalence, Incidence, Prognosis, and Predisposing Conditions for Atrial Fibrillation: Population-Based Estimates, Am J Cardiol 1998;82(8A):2N–9N http://www.sciencedirect.com/scien ce/article/pii/S0002914998005839 [ abstract only ] _____________________________________________________________________________________ • CHADS 2 (Gage) Gage BF. Validation of Clinical Classification Schemes for Predicting Stroke Results From the National Registry of Atrial Fibrillation, JAMA 2001;285(22):2864-2870. [doi:10.1001/jama.285.22.2864] www.excellence.acforum.org/sit es/default/files/Gage.CHADS2_.pdf _____________________________________________________________________________________ 31 _____________________________________________________________________________________ • CHA 2DS2-VASc (Lip et al.) Lip GYH et al . Refining Clinical Risk Stratification fo r Predicting Stroke and Thromboembolism in Atrial Fibrillation Using a Novel Risk Factor-B ased Approach: The Euro Heart Survey on Atrial Fibrillation, CHEST 2010; 137(2):263–272 Lip GYH et al .. Identifying Patients at High Risk for Str oke Despite Anticoagulation: A Comparison of Contemporary Stroke Risk Stratification Schemes in an Anticoagulated Atrial Fibrillation Cohort, Stroke. 2010;41:2731-2738 [doi: 10.1161/STROKEAHA.110.590257 ] http://stroke.ahajournals.o rg/content/41/12/2731.ful l _____________________________________________________________________________________ • Qstroke Hippisley-Cox J et al. Derivation and validation of QStroke score for predicting risk of ischaemic stroke in primary care and comparison with other ri sk scores: a prospective open cohort study, BMJ 2013; 346: f2573. http://www.bmj.com/content/346/bmj.f2573.full.pdf+html _____________________________________________________________________________________ • Study Statistics LaValley MP. Statistical Primer for Cardiovascular Research: Logistic Regression, Circulation 2008; 117: 2395-2399 [ doi: 10.1161/CIRCULATIONAHA.106.682658] http://circ.ahajournals.org/content/117/18/2395.full _____________________________________________________________________________________ • ESC 2012 AFIB Guidelines Camm AJ et al. 2012 focused update of the ESC Guidelines fo r the management of atrial fibrillation: An update of the 2010 ESC Guidelines for th e management of atrial fibrillation, European Heart Journal (2012) 33, 2719–2747 [ doi:10.1093/eurheartj/ehs253 ] http://www.escardio.org/guidelines-surveys/ esc-guidelines/Pages/ atrial-fibrillation.aspx _____________________________________________________________________________________ 32 _____________________________________________________________________________________ • OAC After Catheter Ablation? Levitt HL et al. Is There a Need to Continue Anticoagula tion Following "Successful" Atrial Fibrillation Ablation? Prev. Cardiol . 2009;12:34–37 [doi: 10.1111/j.1751-7141.2008.00013.x] http://www.readcube.com/articles/ 10.1111/j.1751-7141.2008.00013.x See also page 2741 of the previous reference ESC 2012. _____________________________________________________________________________________