Europen Journl of Crdio-thorcic Surgery 20 (2001) 1183 1187 www.elsevier.com/locte/ejcts Evlution of the reltionship between preopertive risk scores, postopertive nd totl length of stys nd hospitl costs in coronry bypss surgery q T.S. Kurki, *,U.Häkkinen b, J. Luhrnt c,j.rämö d, M. Leijl c HUCH, Deprtment of Anesthesi nd Intensive Cre Medicine, Meilhti Hospitl, FIN-00029 HUS, Helsinki, Finlnd b Ntionl Reserch nd Development Centre for Welfre nd Helth, FIN-00530 Helsinki, Finlnd c Helsinki University Centrl Hospitl, FIN-00290 HUS, Helsinki, Finlnd d HUCH, Deprtment of Crdiothorcic Surgery, Meilhti Hospitl, FIN-00029 HUS, Helsinki, Finlnd Received 21 My 2001; received in revised form 20 August 2001; ccepted 31 August 2001 Abstrct Objectives: Severl risk indices hve been developed for the prediction of postopertive mortlity nd morbidity in coronry rtery bypss opertions, in which the risk scores re currently recorded s routine prxis. The im of the present study ws to determine whether the risk scores cn be used to predict the hospitl (LOS) nd postopertive (POS) lengths of sty nd totl costs mong coronry rtery bypss grft (CABG) ptients. Methods: All first-time CABG ptients (n ¼ 2104) treted t Helsinki University Centrl Hospitl during 1997 1998 were preopertively scored using the Clevelnd Clinic preopertive model. A multivrite nlysis ws used to evlute the effects of the risk scores on the LOS nd POS nd totl costs. Results: The men preopertive risk score for the ptients ws 1.69. The increse in preopertive risk score ws ssocited with n increse in the LOS (0.8 dys by point), nd POS (with 0.55 dys by point). An ge over 74 yers incresed the LOS by n extr dy. The men totl cost for the CABG procedure ws 8750 euros (SD 4430 euros). The costs incresed s the risk score incresed. Compred with the zero risk score, score vlue of 2 ws ssocited with 1300 euros increse in totl cost nd score vlue of over 6 ws ssocited with n over 7000 euros cost increse. On verge, the costs incresed by 6980 euros (80%) for one mjor compliction nd by 935 euros (10%) in the elderly (.74 yers of ge). Conclusions: The results show tht incresing risk scores were ssocited with longer postopertive hospitl lengths of sty (POS nd LOS) nd with incresed totl costs. An ge over 74 yers ppers to be n independent risk fctor in incresed POS, LOS nd totl cost. These results my help to estimte the impct of the preopertive risk profile on the resource requirement in CABG surgery. q 2001 Elsevier Science B.V. All rights reserved. Keywords: Coronry bypss surgery; Costs; Length of hospitl sty; Morbidity; Risk scores 1. Introduction q Presented t the Annul Meeting of the Americn Society of Anesthesiologists, Sn Frncisco, CA, October 14 18, 2000. * Corresponding uthor. Lukupolku 19, 00680 Helsinki, Finlnd. Tel.: 1358-9-47161781. E-mil ddress: tuul.kurki@hus.fi (T.S. Kurki). In Western societies, the costs of helthcre re incresing yer by yer [1]. One of the resons for this growth is the fct tht the popultion in Western countries is geing [2]. Elderly ptients (over 70 yers of ge) hve high incidence of chronic diseses such s hypertension, dibetes nd chronic lung disese. Another fctor contributing to incresing costs is the use of expensive technology such s coronry rtery bypss grft (CABG) surgery. The third prty (stte, communities, insurnce compnies nd helth mintennce orgniztions) hs tken n ctive prt in controlling hospitl costs. The requirement is to cut costs without diminishing the qulity of cre. One solution is to increse efficiency; hospitls need to pln their opertions to use vilble resources in n optiml fshion. This is especilly true for CABG surgery which is one of the most expensive elective procedures in hospitl. The cost of CABG surgery cn vry enormously between ptients with uncomplicted recovery nd those who suffer from postopertive complictions such s stroke, rrhythmi nd infections. Prediction of postopertive morbidity would fcilitte decisions to operte, llocte resources nd estimte costs [1]. The purpose of the present investigtion ws to evlute the reltionship between preopertive risk scores (Clevelnd model) [3] nd hospitl length of sty (LOS), postopertive length of sty (POS) s well s totl hospitl costs in Helsinki University Centrl Hospitl. 1010-7940/01/$ - see front mtter q 2001 Elsevier Science B.V. All rights reserved. PII: S1010-7940(01)00988-5
1184 T.S. Kurki et l. / Europen Journl of Crdio-thorcic Surgery 20 (2001) 1183 1187 Tble 1 Clevelnd risk-scoring system Preopertive fctor Age (yers) 65 74 1 $75 2 Chronic obstructive pulmonry disese 1 Renl filure; s-cretinine (mmol/l) 141 167 1 $168 4 Reopertion 3 Cerebrovsculr disese 1 Dibetes, orl or insulin 1 Opertive ortic vlve stenosis 1 Opertive mitrl vlve insufficiency 3 Severe left ventriculr dysfunction 3 Emergency 6 Anemi, hemtocrit,34 1 Prior vsculr surgery 2 Weight,65 kg 1 2. Mteril nd methods Score We studied retrospectively 2104 first-time CABG ptients operted on in Helsinki University Centrl Hospitl during 1997 1998. The men ge of the CABG ptients ws 63.13 yers (SD 9.4, rnge 32 88 yers). There were 1560 men (74.1%) nd 544 women (25.9%). The ptients were preopertively scored in the clinic ccording to the Clevelnd Clinic preopertive model [3], which predicts both mortlity nd morbidity. The cutoff point for morbidity is score level of 4 or more nd for mortlity score level of 6 or more. The mximum possible score in this model is 32. The men preopertive risk score ws 1.69. The preopertive fctors nd score vlues for this model re listed in Tble 1. This model ws used becuse it is ccepted s stndrd risk strtifiction model for ll crdic surgery centres ntionwide. Hospitls collect the preopertive risk score dt for every ptient; the dt were included in the Ntionl Dischrge Register. The ptient level nd risk score dt were collected from the Ntionl Dischrge Register. The register lso includes the following postopertive outcome mesures: stroke, myocrdil infrction, infection (medistinitis, wound infection or pneumoni) nd bleeding (cusing resternotomy). Other complictions include: renl filure requiring dilysis, use of the intr-ortic blloon pump, ventiltor support over 48 h or prolonged sty in the intensive cre unit (ICU) for over 3 dys. Deth during the hospitl sty ws lso recorded by the Ntionl Dischrge Register. In ddition, ptient-level cost dt were linked using personl identifiction numbers. The costs re bsed on the hospitl s ptient-level ccounting system. The ptient-level costs include totl costs (including deprecition) excluding user (inptient) chrges of 40 euros/dy. We clculted men totl costs, LOS nd POS ssocited with different levels of risk scores. In ddition we used multivrite regression nlysis to evlute the effect of risk scores on the three dependent vribles. Thus, it is possible to evlute the effects of risk scores fter controlling for other relevnt vribles. The multivrite regression nlysis ws performed by using three lterntive specifictions of independent vribles. Firstly (model 1), the dependent vribles were regressed ginst vribles describing risk scores only. Secondly (model 2), vrible describing n ge over 74 yers ws included in ddition to risk scores in order to evlute whether the ge of the ptient ws dequtely considered in the risk score. In the third model, vribles describing postopertive complictions (morbidity), deth nd the chrcter of the opertion were included in the nlysis to evlute the sensitivity of the effect of risk fctors fter controlling for other vribles. All explntory vribles were dichotomous. Risk scores were divided into six dichotomous vribles ccording to their distribution. In the regression nlysis the functionl form nd the nture of the dependent vribles were considered. Cost per ptient ws usully nlyzed using liner or log-liner functionl form. By mens of Box Cox trnsformtion nd the likelihood rtio (LR) test it is possible to obtin guidnce for the pproprite functionl form. In this cse both linerity nd log-linerity of the independent vrible were rejected. Insted, the use of vlue of 0.7 for lmbd gve the mximum vlue for log-likelihood. 1 The number of hospitl dys (LOS or POS) cn hve only nonnegtive integer vlues, which mens tht the use of ordinry lest-squres regression is limited. The Poisson distribution is often used to describe discrete vrible nd Poisson regression to explin its vrition. In the present study, s is usul in this type of nlysis, the dt included over dispersion compred with the Poisson distribution. This problem cn often be hndled by negtive binomil regression. The results of multivrite nlysis re illustrted by clculting the mrginl effects of the explntory vribles on the dependent vribles. Since ll vribles re dichotomous, the mrginl effects indicte how chnge in the vrible (from 0 to 1) chnges the vlue of the dependent vrible. In the cse of risk score vribles the reference vlue is risk score vlue of zero, which mens tht the mrginl effect compres the effects of different risk scores with the sitution in which the risk score is zero. 2.1. Ethicl issues This study did not interfere with the tretment of ptients, nd the retrospective dtbse ws orgnized in wy tht mkes the identifiction of n individul ptient impossible. 1 We estimted the optimum vlue for lmbd for ech of three lterntive models. In ech cse the vlue 0.7 gve the highest log-likelihood rtio. In prctice this mens tht we performed the following trnsformtion to the cost vrible (c): (c 0.7 2 1)/0.7.
T.S. Kurki et l. / Europen Journl of Crdio-thorcic Surgery 20 (2001) 1183 1187 1185 Tble 2 Men vlues of dependent vribles t different level risk scores nd their mrginl effects in euros Risk score Cost/dmission (euros) LOS POS Men Mrginl effect Men Mrginl effect Men Mrginl effect 0 7856 0 8.3 0 6.8 0 1 8031 375 8.9 0.6 7.5 0.7 2 9036 1380 9.7 1.4 8.1 1.3 3 9336 1680 10.3 2.0 8.6 1.8 4 6 10205 2549 10.0 1.7 8.5 1.7.6 14995 7339 11.3 3.5 9.8 3 Comprison to reference vlue (risk score ¼ 0). 3. Results The mjority of the ptients, 1926 (91.5%) out of 2104, hd fvourble, uncomplicted outcomes. Mjor postopertive complictions were documented in 7.3% (173) of ptients. A totl of 25 ptients (1.2%) died during their hospitl sty. Cerebrl complictions (stroke) were the most frequent complictions (46/2104, 2.2%). Postopertive excessive bleeding requiring resternotomy occurred in 29 ptients (1.4%). There were 29 postopertive infections (1.4%). Myocrdil infrction ws documented in 28 ptients (1.3%) postopertively. Other crdic complictions occurred in eight ptients, nd complictions in other orgn systems were documented in ten ptients including renl complictions in three ptients. The ptients showed men LOS vlue of 9.2 dys (Tble 2). With risk score of zero, the men LOS ws 8.3 dys, with risk scores of 4 6 the men LOS ws 10 dys nd with scores of 7 10 the LOS ws 11.3 dys. With risk score over 10 the LOS incresed up to 15 dys (men). The men POS vlue ws 7.7 dys. In ptients with no risk fctors, the POS ws 6.8 dys nd in ptients with scores of.6, the POS ws men of 9.8 dys. The results of negtive binomil regression confirmed the Tble 3 Results of negtive binomil regression models for LOS Risk score b 1 0.6*** c 0.6 0.6* 2 1.4*** 0.9*** 1.0*** 3 1.9*** 1.6*** 1.3*** 4 6 1.7*** 1.4*** 1.2***.6 2.8*** 2.4*** 2.6*** Age over 74 yers 1.4*** 1.0*** Deth 25.8*** Revision opertion 3.9*** Infection 8.9*** Stroke 3.1*** Myocrdil infrction 2.7*** Other complictions 5.9*** The estimted coefficients re illustrted by clculting their mrginl effects in hospitl dys. b Comprison to reference vlue (risk score ¼ 0). c *P, 0:05, **P, 0:01, ***P, 0:001. results given by mens (Tble 3). For exmple, the mrginl effect of risk score of 1 with one vrible is 0.6, i.e. ptients hving risk score of 1 will hve bout 0.6 dys longer length of sty compred with the reference vlue (risk score ¼ 0). The effect of risk scores will be rther stble even fter controlling for ge nd compliction vribles. However, ptients over 74 yers of ge tend to hve bout 1.4 dys longer length of sty compred with younger ptients. The results of negtive binomil regression (Tble 4) of POS vlues were rther similr to those of LOS vlues. The hospitl costs were on verge 8750 euros per ptient, nd they incresed in reltion to the risk score. According to the results of Box Cox regression (Tble 5), ptient with risk score of 1 ws ssocited with bout 350 euros higher cost compred with the reference vlue (risk score ¼ 0). With risk score of 4, the costs were incresed by 2500 euros compred with zero risk score level. If the risk score ws over 6, the dditionl cost ws 7000 euros. The verge dditionl cost for the elderly ws 935 euros. Postopertive complictions incresed the totl cost on verge by 6700 euros per compliction (rnge 4200 9700 euros). Reopertion due to excessive bleeding Tble 4 Results of negtive binomil regression models for POS Risk score b 1 0.7*** c 0.7*** 0.7*** 2 1.3*** 0.9*** 1.0*** 3 1.8*** 1.5*** 1.3*** 4 6 1.6*** 1.4*** 1.2***.6 2.8*** 2.5*** 2.6*** Age over 74 yers 1.2*** 0.9*** Deth 26.5*** Revision opertion 3.8*** Infection 8.3*** Stroke 2.6*** Myocrdil infrction 2.9*** Other complictions 5.6*** The estimted coefficients re illustrted by clculting their mrginl effects in postopertive hospitl dys. b Comprison to reference vlue (risk score ¼ 0). c *P, 0:05, **P, 0:01, ***P, 0:001.
1186 T.S. Kurki et l. / Europen Journl of Crdio-thorcic Surgery 20 (2001) 1183 1187 Tble 5 The results of Box Cox regression models for cost/dmission Risk score 1 345 354 286 2 1300*** b 1019*** 915*** 3 c 1582*** 1367*** 1022*** 4 6 c 2447*** 2224*** 1933***.6 c 7008*** 6719*** 6142*** Age over 74 yers 935*** 559*** Deth 3864*** Revision opertion 9789*** Infection 6228*** Stroke 4641*** Myocrdil infrction 4309*** Other complictions 8423*** The estimted coefficients re illustrted by clculting their effects in monetry terms (euros). This indictes how much the chnge in the vrible will chnge the cost per ptient. b *P, 0:05, **P, 0:01, ***P, 0:001. c Comprison to reference vlue (risk score ¼ 0). The impct of the individul risk fctors ws not investigted since, with the exception of ge, we did not hve ccess to single risk fctors. In elderly ptients (over 75 yers of ge), the length of hospitl sty ws incresed by n verge of 1.4 dys compred with younger ptients. The totl hospitl cost ws incresed by 940 euros (men) in the elderly, even fter controlling for other risk fctors. This result is interesting, considering the fct tht the popultion in Western countries is geing. Furthermore, the epidemiology of coronry rtery disese suggests tht the disese will become symptomtic t lter ge in the future. Advnced ge is ssocited with generl rteriosclerosis, ffecting not only coronry rteries but lso the renl nd cerebrl rteries. Elderly ptients (over 70 yers of ge) lso hve higher incidence of hypertension, dibetes nd chronic lung disese. Nevertheless, it is surprising tht ge lone hs such mjor impct on costs in CABG surgery. Further prospective studies re needed to evlute the effects of single preopertive comorbidity fctors on LOS vlues nd totl costs since their reltive impct on ws the most expensive postopertive compliction (n dditionl cost of 9760 euros). Infections nd other complictions incresed the bsic totl cost by 6200 nd 8400 euros, respectively. 4. Discussion Severl risk indexes hve been developed for the prediction of postopertive mortlity nd morbidity in coronry bypss surgery [4 8]. Although the risk indexes cn only give rough estimte of the risk for n individul ptient, they cn be used for plnning purposes t the popultion level. The results of the present study demonstrte tht there is close reltionship between the preopertive risk scores s mesured by the Clevelnd model on one hnd, nd postopertive nd totl lengths of sty nd totl cost on the other hnd. These results extend the previous findings tht suggested reltionship between LOS nd preopertive ptient chrcteristics [9]. Risk strtifiction models will id hospitls in predicting the need for resources. In the present study we used the Clevelnd model becuse it is routinely used ntionwide nd is collected from the Ntionl Dischrge Register. Most risk fctors in the Clevelnd model re the sme in the Prsonnet nd EuroSCORE models (Tble 6), lthough the weights my vry. These models re derived using logistic regression nlysis. The Clevelnd model predicts both morbidity nd mortlity, wheres the EuroSCORE nd Prsonnet models hve been vlidted only for mortlity [10]. Our intention ws to investigte whether the risk scores recorded routinely for other purposes could lso be used to predict the LOS, POS nd economic outcome. Our im ws neither to develop new score for the economic outcome, nor investigte the impct of individul vribles on costs. Tble 6 Comprison of most importnt risk fctors in Clevelnd, EuroSCORE nd Prsonnet models Clevelnd model EuroSCORE Prsonnet Ptient-relted fctors Age x x x Gender Femle Chronic obstructive x x x pulmonry disese/ sthm Obesity/weight,65 kg BMI.35 Renl dysfunction x x x Previous crdic surgery x x x Neurologicl dysfunction x x x (crdiovsculr disese) Dibetes x x Criticl preopertive x x stte Intr-ortic blloon pump x x Crdic-relted fctors Unstble ngin pectoris x Active endocrditis x Hypertension x Left ventriculr x x x dysfunction Recent myocrdil x infrction Pulmonry hypertension x Prior vsculr surgery x x Opertion-relted fctors Emergency x x x Combined surgery x x x Surgery of thorcic ort x Post infrct septl rupture x Left ventriculr neurysm x Anemi x
T.S. Kurki et l. / Europen Journl of Crdio-thorcic Surgery 20 (2001) 1183 1187 1187 costs my be different from their weight in the estblished risk score. Other risk-scoring systems should lso be tested to evlute the rel benefits of risk strtifiction for resource lloction nd in long-term plnning of hospitl schedules in crdic surgery. In conclusion, modelling of hospitl costs nd prediction of length of sty is possible on the bsis of preopertive risk scores. This lso mkes it possible to llocte resources nd to pln weekly schedules for CABG opertions. However, the most obvious use of the present results is in long-term plnning. Risk strtifiction my help decision-mkers to estimte the need for resources for these different ptient groups in the future. References [1] Ferrris VA, Ferrris SP, Singh A. Opertive outcome nd hospitl cost. J Thorc Crdiovsc Surg 1998;115:593 603. [2] Weintrub WS, Crver JM, Jones EL, Gott JP. Improving cost nd outcome of coronry surgery. Circultion 1998;II:23 28. [3] Higgins TL, Estfnous FG, Loop FD, Beck GJ, Blum JM, Prnndi L. Strtifiction of morbidity nd mortlity outcome by preopertive risk fctors in coronry rtery bypss ptients. J Am Med Assoc 1992;267:2344 2348. [4] Prsonnet V, Den D, Bernstein AD. A method of uniform strtifiction of risk for evluting the results of surgery in cquired dult hert disese. Circultion 1989;79(1):3 12. [5] Kurki TSO, Ktj M. Preopertive prediction of postopertive morbidity in coronry rtery bypss grfting. Ann Thorc Surg 1996;61:1740 1745. [6] Nshef SAM, Roques F, Michel P, Guducheu E, Lemeshow S, Slmon R, the EuroSCORE study group. Europen System for Crdic Opertive Risk Evlution (EuroSCORE). Eur J Crdiothorc Surg 1999;16:9 13. [7] Egle KA, Guyton RA, Dvidoff R. ACC/AHA guidelines for coronry bypss grft surgery. J Am Coll Crdiol 1999;34:1262 1347. [8] Tu JV, Jgll SB, Nylor CD. Multicenter vlidtion of risk index for mortlity, intensive cre unit sty nd overll hospitl length of sty fter crdic surgery. Circultion 1995;91:677 684. [9] Ghli WA, Hll RE, Ash AS, Moskowitz MA. Identifying pre- nd postopertive predictors of cost nd length of sty for coronry rtery bypss surgery. Am J Med Qul 1999;14(6):248 254. [10] Geissler HJ, Hölzl P, Mrohl S, Kuhn-Regnier F, Mehlhorn U, Sudkmp M, de Vivie ER. Risk strtifiction in hert surgery: comprison of six score systems. Eur J Crdiothorc Surg 2000;17:400 406.