DEVELOPING A DECISION SUPPORT SYSTEM FOR NURSE SCHEDULING AT A PUBLIC HOSPITAL IN SOUTH AFRICA

Transcription

1 CIE42 Proceedngs, 6-8 July 202, Cape Town, outh Afrca 202 CIE & AIIE DEVELOPING A DECIION UPPORT YTEM FOR NURE CHEDULING AT A PUBLIC HOPITAL IN OUTH AFRICA. Fredrch * and L. van Dyk Department of Industral Engneerng Unversty of tellenbosch, outh Afrca lvd@sun.ac.za ABTRACT tellenbosch Hosptal s stuated n the Western Cape of outh Afrca. It s an 85- bed publc faclty and serves a communty of people. Based on a root cause analyss the concluson was made that nurse utlzaton and employee morale at the hosptal could be mproved by grantng nurses more control over schedules. The purpose of ths project s to develop a nurse schedulng Decson upport ystem (D). Ths system allows nurses to be more nvolved n the annual day-and-nght shft schedulng and ward preferences, as these factors have the most sgnfcant mpact on nurses at work. The day-and-nght schedule s solved wth an nteger programmng model. The results are utlzed as an nput to the custom-made algorthm, whch mtates the thought process, of the unt managers schedulng process, n arrvng at a basc feasble soluton for ward assgnments. A flowchart supportng the documentaton and understandng of the code has been developed along wth a testng table to verfy the results of the code. Fnally, a user valdaton of the D at tellenbosch Hosptal confrms ts usefulness and effectveness to support nurse schedulng decson makng, enhance nurse utlzaton and mprove employee morale by ncludng nurses n the schedulng process. * Correspondng Author 5-

2 CIE42 Proceedngs, 6-8 July 202, Cape Town, outh Afrca 202 CIE & AIIE INTRODUCTION The provson of health care to all ts ctzens s of great concern to any government, outh Afrca s no excepton. One of the goals of the outh Afrca government s to create a long and healthy lfe for all outh Afrcans []. The outh Afrcan health care system faces several challenges whch make health care mprovement dffcult. Accordng to Harrson [2], one of the central concerns les n the management of health systems. tellenbosch Hosptal s a typcal dstrct hosptal, founded n 942. It s an 85-bed, nonproft publc faclty that serves a communty of people. Approxmately 5000 patents are treated monthly, comprsng an equal number of outpatents and npatents. The government subsdzes tellenbosch Hosptal wth about R95 mllon annually n an attempt to reach natonal goals for the health sector. Prmary health care, at no cost to all ctzens, s the ultmate aspraton of a government hosptal. The nurses are the provders of ths servce and they ensure the qualty of care [3]. Health care and nursng servces were thus dentfed as the focal pont for ths project. A cause-and-effect analyss (Error! Reference source not found.) was executed to determne the root causes of the problems at tellenbosch Hosptal. Consderng Harrson s [2]work, as well as personal observatons, the qualty of health care and nursng servces at tellenbosch Hosptal s affected by four major causes. These causes are (Error! Reference source not found.- man bones):. the lack of authorty of management, 2. low employee morale, 3. lmted and nept management of fnances and general neffectveness and 4.absenteesm of staff. Utlzng ndustral engneerng tools to upgrade the nurse schedulng process yelds potental opportuntes for mprovng the qualty of health care and nursng servces at tellenbosch Hosptal, and possbly other health care facltes n outh Afrca. 2 PURPOE AND METHODOLOGY The most common root causes of substandard health care dentfed at tellenbosch Hosptal are; the complexty of schedulng and ncorporatng nurses preferences n schedules, nadequate nurse utlzaton and shft coverage, and nurses lack of authorty and control over schedules. Resolvng these root causes could ultmately allevate the man predcament of nsuffcent qualty of health care and nursng servces at tellenbosch Hosptal. The purpose of ths project s to develop a nurse schedulng Decson upport ystem (D). The followng objectves are pursued to acheve ths purpose: Provde decson makers wth a D, Improve and enhance nurse utlzaton, Involve nurses n the schedulng process and Improve employee morale. The methodology s a combnaton of the strateges developed by Turban et al[4]and Wnston [], as follows: Decde on a scope of decsons that need to be ntegrated nto the programme n order to ncorporate the most relevant factors, Observe the system and decde on whch data to use, Decde whch development software to use, Choose an algorthm, Defne a mathematcal model that descrbes the real-world problem and Fnd possble solutons to the modelled problem and evaluate them. 3 THE DECIION UPPORT YTEM 5-2

3 CIE42 Proceedngs, 6-8 July 202, Cape Town, outh Afrca 202 CIE & AIIE Decson upport ystems couple the ntellectual resources of ndvduals wth the capabltes of the computer to mprove the qualty of decsons. It s a computer-based support system for management decson makers who deal wth sem-structured problems [4]. The most mportant D nputs for tellenbosch Hosptal were dentfed as; the dfferent types of nurses (sster, staff nurse, and assstant nurse), nurses preferences, nurses avalablty, day or nght shft assgnment, nurse requrements and nurses qualfcatons. The process was the schedulng process, whch would transform the nput nformaton nto a feasble assgnment of nurses, and thus form the output. The decson makers are the unt managers who ensure the feasblty of the schedule and gve comments on how t may be mproved. Ths s transferred as nputs and an mproved schedule s then created. Furthermore, due to the fact that the publc has no nsght nto schedulng procedures, the hosptal schedulng process can be classfed as a closed system. 3. Nurse schedulng algorthms The nurse schedulng problem s sem-structured, as t requres human judgment combned wth structured elements, or known varables, for nstance the number of wards. The schedulng problem furthermore supports operatonal actvtes whch call for regular decson makng. A D s a possble soluton to ths type of problem[4]. Nurse schedulng models have been of nterest to operatons researchers for decades, for example Baly s [5]shft personnel schedulng model, whch focuses manly on the day-to-day schedulng. Ths nvolves work patterns, off days, weghng nurse preference aganst cost (whch ncludes patent nconvenence due to understaffng) and takng dfferent constrants nto account, for nstance maxmum day shfts allowed per week. Ths model takes the dayand-nght assgnments and ward assgnments as gven nputs. Ozkarahan[6] uses Baley s [5]work and expands on t. He allocates optmum work patterns to ndvdual nurses based on ther desres and compatbltes. In contrast to Baley s work, Ozkarahan s model takes note of the ward/unt allocatons n the day-to-day schedulng system. Hs model assgns a nurse to a specfc work pattern n a specfc ward. Ackeln and Dowsland[7] on the other hand take day-and-nght schedulng nto account, but gnore nurse type and applcable ward assgnments. Nurse schedulng s also addressed locally, as mentoned n Bester et al[8] n whch a Nurse Rosterng Decson upport ystem (NUROD) was developed for tkland Hosptal, n the Western Cape, usng metaheurstcs. mlar to Baley s [5]model, ths D takes day-andnght, as well as the ward assgnments as gven nputs. Numerous models assume day-and-nght shft, as well as ward assgnments, as gven nputs. Other models only cater for day-and-nght schedulng, and further models address ward assgnments but gnore day-and-nght schedulng. Incorporatng all factors nto one algorthm makes the problem very complex. Consequently, n ths project the D caters for these neglected but mportant schedulng areas that could be used as a possble nput to the other models descrbed. 5-3

4 CIE42 Proceedngs, 6-8 July 202, Cape Town, outh Afrca 202 CIE & AIIE 4.2. Inputs and outputs to the system In the current schedulng system, unt managers meet once a month to manually schedule day-and-nght, as well as ward assgnments, of dfferent nurses wth dverse sklls, to seven wards. Ths s the bulk of the work. ubsequently, shfts are recorded n a shft book accordng to the fxed; three on-, three off-, two on-, two off- days shft pattern. One shft s 2 hours from 7 pm to 7 am or from 7 am to 7 pm. Nurses are always scheduled n pars, thus the second nurse follows a pattern; three off-, three on-, two off-, and two on- days. The unt managers call t the exchange -shft, ths partcular pattern was developed by tellenbosch Hosptal. Accordng to the unt manager, there s no specfc reason for ths pattern except that t reduces complexty of schedulng. The fve unt managers themselves and some of the assstant nurses follow a stretch shft whch s a normal workng day shft from 7 am to 4 pm from Monday to Frday. The unt managers decde on the shft type for each nurse. Day-and-nght shft assgnments are currently done each month, and each nurse has to work three consecutve months of nght shfts. Ths arrangement s part of hosptal regulatons. Currently, no nput from nurses s gven and nurses do not know when they have to commence nght shfts. In the D, day-and-nght shft assgnments are assgned on an annual bass. As the day-and-nght schedules have a crucal mpact on the nurses workng lves, ther preference s used as an nput when creatng a schedule. Nurses are able to select specfc quarters of the year durng whch they prefer to work nght shfts. In terms of the monthly ward assgnments, unt managers gve ther preferences durng the current quarter. Nurse herarchy s a parameter used by the D to dentfy nurse qualfcatons. It conssts of four levels, wth unt managers at the hghest level. The second level conssts of the ssters (a term used for hgher qualfed nurses), followed by staff nurses and then assstant nurses. The nurse herarchy s recorded as a fxed nput nto the D. Unt managers enter the names of the nurses to be assgned, as well as the herarchy status of each nurse. hft patterns of unt managers are fxed and not ward specfc, therefore unt managers are not ncluded n the schedulng system. Nurses are ranked and promoted accordng to tranng and experence. Currently the hosptal has 24 ssters, 37 assstant nurses and 23 staff nurses, all of whom have varyng sklls and avalablty whch changes over tme. Up to ten addtonal nurses per nurse type may be added, makng the model moderately flexble(?). In the case of 5-4

5 CIE42 Proceedngs, 6-8 July 202, Cape Town, outh Afrca 202 CIE & AIIE tellenbosch Hosptal, nurse numbers are not lkely to ncrease sgnfcantly due to the restrcted budget. Fgure : Inputs and outputs to the decson support system In the D, the number of ssters, staff nurses as well as assstant nurses, and ther sklls requred per ward, are taken as a varable nput by unt managers. Ward assgnments are done once a month as ndcated by the tme trgger n Error! Reference source not found.. Nurses have the opton to gve a preferred ward assgnment nput. The Hosptal has seven wards: the Pedatrc ward, Ward A (chronc patents), Ward B (women), Ward C (men), Theatre, the Trauma/Accdents ward and the Maternty ward. Each of these wards needs a mx of ssters, assstant nurses and staff nurses as specfed by the unt managers. The mx, as well as the avalablty of nurses, s a varable nput suppled by unt managers. Nurses who are on study-, normal-, maternty-or sck leave durng a schedulng month may be hghlghted(?). For the D, the wards are a fxed nput, as t s unlkely that an addtonal ward wll be added to the hosptal. Nursng costs per shft are not ncluded as an nput, because all the nurses are usually assgned, and as ndcated n the cause-and-effect dagram, hgh nursng costs are a result of nsuffcent schedules, absenteesm and resultng overtme. The D s objectve s to mprove schedules and nvolve nurses, and thus reduce costs due to a more effcent approach and soluton. Patent admsson data s not drectly ncorporated n the system but they do follow clear trends whch the experenced unt managers are aware of. For nstance, admssons ncrease at the end of the month after people have receved ther salares, as well as on Frday nghts (Lnders& kppers, personal communcaton, 3 May 20). In the D, unt managers manually adjust the number of nurses requred per month per ward, and assgn extra nurses on daly shfts as requred. Ths practce makes patent forecastng redundant. The basc feasble soluton, as ndcated as an output of the D process n Error! Reference source not found., s smlar to the current output report of the manual nurse schedulng process. Ths s done so that the unt managers may adapt more comfortably to the new schedulng method. The D s output sheet ncludes the dfferent ward assgnments, dayand-nght assgnments and dentfes nurses who are not avalable for perods of the month. Unt managers manually document, n the output sheet, the reasons why nurses are not avalable. Furthermore, nurse pool sectons are added, whch ndcate the nurses that have not been assgned for varous reasons. Based on these facts, unt managers apply ther own judgment to the output sheet, adjustng the computer generated day-and-nght schedule and ward assgnments, as to obtan a good basc feasble soluton. 4 LINEAR PROGRAMME FOR NURE CHEDULING AT TELLENBOCH HOPITAL Lnear programmng and a custom-made algorthm were determned to be the best optons accordng to the Analytcal Herarchy Process (AHP). Consequently, these methods were used to develop the day-and-nght assgnment, as well as the ward assgnments. Although metaheurstcs s a strong contender, the approach was not pursued for the purposes of ths project. However, a study that was done parallel to ths project,used a genetc algorthm (metaheurstc) to solve a schedulng problem, specfcally adapted for tellenbosch Hosptal [9]). 5.. Day-and-Nght Integer Programmng Model The day-and-nght schedulng s separated nto three parts: sster schedulng, assstant nurse schedulng and staff nurse schedulng. Each one of these s solved n the same manner. The frst constrant s mplemented n the structure of the model. Nurses have to work at least three consecutve months of nght shfts per year. Thus, the year s separated nto four quarters. 5-5

6 CIE42 Proceedngs, 6-8 July 202, Cape Town, outh Afrca 202 CIE & AIIE If the objectve functon s maxmzed, the system assgns more shfts than needed by the shft coverage constrants. Consequently, n ths model the objectve functon mnmzes the dssatsfacton of nurses. J j = Nurse preference for quarter j. J s a matrx wth four columns for four nurses and a number of rows respectve to the number of nurses that need to be assgned to a nght shft. Nurses may choose J j as 0 (most preferred quarter for nght shft), 5 (f nurses do not care to work nght shfts durng a specfc quarter) or 0 (f nurses do not want to work nght shfts durng a specfc quarter). j = Assgned shft of nurse for quarter j. s a matrx of ones (assgned shft) and zeros (no shft) wth four columns for four nurses and a number of rows respectve to the number of nurses that need to be assgned to a nght shft. R = Coverage - nursng requrements for three months to cover needs as specfed by unt managers. It s assumed that the basc nght shft nursng requrements are the same each quarter. Ths s realstc as wards are assgned fxed numbers of each nurse type each day. mn z = j n= m= J j j () ubject to: Nursng requrements to be satsfed each quarter: n= n= n= n= R Quarter R Quarter 2 R Quarter3 R Quarter 4 (2) Mnmum shfts to be assgned per nurse per year, where each nurse has to work at least one quarter: 2 Nurse (3) Nurse2 Nurse Where: R 0, Jj = 0,5or0and j are ntegers 5-6

7 CIE42 Proceedngs, 6-8 July 202, Cape Town, outh Afrca 202 CIE & AIIE The model was solved usng Excel olver. Theoretcally the nteger programmng model should converge to the optmal soluton but, due to the complexty of the model, and computatonal constrants of olver, t was necessary to apply a roundng heurstc. Consequently, the soluton s not necessarly optmal. Nonetheless, comparng the soluton to the nurses preferences showed satsfactory results of up to 90% of the cells postvely correspondng to nurses preferences and the coverage constrant was also adhered to. Unt managers may adjust the schedule by shftng assgnments. All three models reacted smlarly and t could thus be sad that the objectve of creatng a pont of departure for the day-and-nght schedule has been acheved. Lmtatons of ths model are as mentoned, the necessty of a roundng heurstc, whch leads to suboptmal solutons. Addtonally, ths model s rgd as nurses may only choose dfferent quarters f ther months are pre-set. Fnally, solvng wth Excel olver lmts the sze of the model. It cannot work wth more than 200 varables and thus, the problem cannot be solved for consderably larger models than the one presented, except f dfferent software s used, whch usually means more cost. 5 YTEM VALIDATION The D valdaton underwent several teratons durng ts development phase. The system was tested wth the users and then taken back to ntroduce mprovements. 6.. Fnal user valdaton The fnal user valdaton was performed on 9/09/20 wth actual nurse data for the month. The staff manager and two unt managers were present. Although these users were sceptcal about the outcomes and the use of the D durng ts development phase, the feedback receved about the latest verson of the model was postve and enthusastc. The ntervew was conducted n Afrkaans, the language whch the unt managers were most comfortable wth, n order to get optmal feedback. They were asked the followng predesgn questons: 6... Is the D useful to you? The unt managers stated that the D s useful to them, as t performs the same computatons they carry out n a tme consumng process, n a few seconds. They sad that startng from an already pre-solved problem speeds up the schedulng process consderably and releves them of the tedous teratons of preparng the schedule, whch often results n nadequate schedules and underutlzed nurses. Furthermore, arguments among unt managers over where specfc resources should be allocated could be prevented. In ther opnon the system supports unt manager s decson makng, wthout compromsng control Does D solve the problem? Accordng to the unt managers, the D does solve problems for them at the hosptal. They sad that documentaton of all the nurses, the Unassgned Pool and the Unavalable Pool ads them n enhancng nurse utlzaton. Unt managers, wth the help of D, have an overvew of all staff and ther attrbutes, and they can thus assgn them n an optmal way. Ths ultmately reduces overtme for nurses, as unt managers can plan schedules n advance. chedulng the nght shfts one year n advance provdes farer schedules and presents nurses wth the opportunty to be aware of upcomng nght shft perods, so that they may organze prvate matters n advance. A more convenent schedule for nurses wll mprove employee morale and reduce absenteesm n the future Do you understand how ths D works? The D and ts basc workngs were explaned to the ssters. As the system emulates the nurses schedulng process, they understood how the system works. They were pleased that ther nput was used so extensvely n the development of the D. 5-7

8 CIE42 Proceedngs, 6-8 July 202, Cape Town, outh Afrca 202 CIE & AIIE Is the D user-frendly? The nurses understood the process of executng the D. They enjoyed the colourful desgn of the programme and apprecated the data valdaton functon whch prevents them from enterng nvald data. However, all of them, beng begnners wth handlng computers, requested a tranng sesson n usng the D and especally olver. 6 CONCLUION In ths project a problem at tellenbosch Hosptal was solved by defnng the problem and developng and valdatng the soluton wth the nput of ts future users. More specfcally, the purpose of ths project was to resolve root causes at a specfc publc hosptal (name removed for peer revew purposes) by desgnng a D related to the schedulng of nursng resources. The followng objectves were set to acheve the purpose of the project: Provde decson makers wth a D, Improve and enhance nurse utlzaton, Involve nurses n schedulng and Improve employee morale. The system has not been mplemented yet, and the mpact on employee morale and absenteesm rates was not quantfed. Nevertheless, unt managers stated that ncludng nurses n the schedulng process s of sgnfcant mportance to the nurses, as knowng ther day-and-nght schedule well n advance enables them to plan ther personal lves accordng to ther work shfts and thus attend work more regularly. Ths potentally mnmzes overtme for nurses and reduces frcton wth management. Nurses are allowed to gan control over ther schedules, and consequently a more comfortable work envronment may be created, whch wll potentally lead to mproved employee morale. These results can be extrapolated to other health care nsttutons n outh Afrca. 7 REFERENCE [] Wnston, W.L., and J. B. Goldberg, Operatons research: applcatons and algorthms, Duxbury press, [2] Presdency of.a., Medum Term trategc Framework, July [Onlne]. Avalable: [Accessed 0 Oct 20]. [3] Harrson, D. An Overvew of Health and Healtcare n outh Afrca: Prortes, Progress and Prospects for New Gans, 24 January 200. [Onlne]. Avalable: [Accessed eptember 20]. [4] Turban, E., Aronson, J., and Lang, T., Decson upport ystems and Intellgent ystems, Upper addle Rver, New Jersey: Pearson Educaton LTD, [5] Baly, J., Integrated days off and shft personnel schedulng, Computers & Industral Engneerng, vol. 9, no. 4, pp , 985. [6] Ozkarahan, I., A dsaggregaton model of a flexble nurse schedulng support system, 5-8

9 CIE42 Proceedngs, 6-8 July 202, Cape Town, outh Afrca 202 CIE & AIIE oco-economc plannng scences, vol. 25, no., pp. 9-26, 99. [7] Burke, E.K., De Causmaecker, P., Van Berghe. G.V., and Van Landeghem, H. The state of the art of nurse rosterng, Journal of schedulng, vol. 7, no. 6, pp , [8] Bester, M.J., Newoudt, I., and Van Vuuren, J.H., Fndng good nurse duty schedules: a case study, Journal of chedulng, vol. 0, no. 6, pp , [9] Treurncht, M.J., Lane-Vsser, T.E., Van Dyk, L., and Fredrch,., A Rosterng Algorthm for a Dstrct Hosptal n outh Afrca, n Annual Conference of the Operatons Research ocety of outh Afrca (ORA), Zmbabwe, 20. [0] Ackeln, U., and Dowsland, K.A., An ndrect genetc algorthm for a nurse-schedulng problem, Computers & Operatons Research, vol. 3, no. 5, pp ,