A STOCHASTIC APPROACH TO NURSE STAFFING AND SCHEDULING PROBLEMS Presented by Sera Kahruman & Elif Ilke Gokce Texas A&M University INEN 689-602 Outline z z Problem definition Nurse staffing problem – – – z Literature and Formulation Computational Study Conclusion and Future study Nurse scheduling problem – – – Formulation Computational Study Conclusion and Future study Problem Definition z z z z Patients in a hospital today are in need of highly skilled nursing care In order for this care to be provided in a timely manner, and meet rigorous quality standards, the right number and type of nursing staff must be available when needed demand is unknown health institutions have to provide service 24 hours a day over seven days a week. Problem Definition z Poor staffing levels lead to: – – – – – Burnout Dissatisfaction Desire to leave the job High cost of staff replacement More patient complications z – Injuries Low quality of service Nurse Skill Types There are 3 types of nurses: 1. Registered Nurses (RNs) 2. Licensed vocational or licensed practical nurses (LVNs,LPNs) 3. Nurse aides z To solve this problem, we need to consider two levels: – Strategic level : determining the long-term regular time nursing levels---staffing – Tactical level : Daily work/shift assignments of the nurses Outline z z Problem definition Nurse staffing problem – – – z Formulation Computational Study Conclusion and Future study Nurse scheduling problem – – – Formulation Computational Study Conclusion and Future study Nurse staffing problem z z z z Hospital administrators determine the regular-time nurse levels annually So, our model should determine the levels of RNs, LPNs(LVNs) and nurse aides for a year Demand is unknown!!! Use 2-stage recourse model Literature z z z Tien and Kamiyama (1982), present a list of personnel scheduling algorithms, which are not restricted to healthcare. Bradley and Martin (1990), distinguish three basic decisions in hospital personnel scheduling: staffing, personnel scheduling and allocation Siferd and Benton (1992), present an excellent review of the factors influencing hospital staffing and scheduling in the United States Literature z Kao, Queyranne(1985) : Budgeting Costs of Nursing In Hospital – – Presents 8 different models Probabilistic or deterministic, different skill types or aggregate skill, single period or multi-period Formulation z Major assumption: There are many medical units within a hospital. We assume that each medical unit decides its own workforce levels and we do not allow inter-unit working schedules. The model is same for all units, except for different demand patterns, and parameters. Formulation z First stage parameters: – – – – ci = total expenses of the hospital for a nurse of type i for the whole year he = Effective working hours of a nurse for one year ADi = minimum annual demand for each skill class nursing hours (can also be a bound specified by hospital regulations) r1,i =Ratio defining the relationship of skill class 1 and i=2,3 (skill class 1 supervises the other types. So there has to be a constraint to make sure that there are enough supervisors) Formulation z First stage decision variables: – – xi : number of regular time nurses of skill type I where i=1,2,3 txi :number of nursing hours of skill type i and higher level skill types which can be used for the demand of lower skill nursing hours i=1,2. Formulation z Second stage parameters: – – – – – oi = cost per overtime hour of a nurse of type i ai = cost per hour of agency nursing providing a service of skill type i ei,t = percentage of total working time of a nurse of skill class i available on month t {due to vacations and etc , effective nursing hours can change from month to month} Oi,t = ratio of the maximum allowed overtime to the effective regular working hours for a nurse of skill type i, on month t di,t = demand realization in hours for nursing skill of type i on month t. Formulation z Second stage decision variables: – – – yi,t= number of overtime hours that nurses of skill type i work on month t. zi,t= number of agency nursing hours of skill type i used for month t. tri :number of nursing hours of skill type i and higher level skill types which can be used for the demand of lower skill nursing hours i=1,2. Formulation min s.t. ~ )} c * x + E{f(x, w ∑i=1 i i 3 he * x1 - tx1 ≥ AD1 he * x 2 + tx1 - tx 2 ≥ AD2 he * x3 + tx 2 ≥ AD3 x1 − r1, 2 * x 2 ≤ 0 x1 − r1,3 * x3 ≤ 0 x1 , x 2 , x3 ≥ 0, integer tx1 , tx 2 ≥ 0 Formulation where for a realization of w, f(x,ŵ) is defined as: 3 12 i =1 t =1 ~ ) = min ( (oi * y + a * z )) f(x, w ∑∑ i i, t i, t s.t. e1,t * x1 + y1,t + z1,t − tr1 ≥ d 1,t ∀t = 1,2....,12 e2,t * x 2 + y 2,t + z 2,t + tr1 − tr2 ≥ d 2,t ∀t = 1,2.....,12 e3,t * x3 + y 3,t + z 3,t + tr2 ≥ d 3,t ∀t = 1,2.....,12 y i, t − xi ,t * Oi ,t ≤ 0 all variables ≥ 0 ∀i = 1,2,3 and t = 1,2,...,12 Computational Study z z z z Use L-shaped algorithm where master problem is solved as a MIP. Comparison with expected value solution, scenario based solution and worst case solution Comparison with linear relaxation of master MIP Observe the effect of demand fluctuation Conclusion and Future Study z z a decision making procedure which considers all medical units at the same time allowing staff movements from one unit to the other one as needed. some parameters in the model are much likely to be probabilistic which will make the model more realistic Outline z z Problem definition Nurse staffing problem – – – z Formulation Computational Study Conclusion and Future study Nurse scheduling problem – – – Formulation Computational Study Conclusion and Future study INTRODUCTION z z z Every hospital needs to repeatedly produce duty rosters for its nursing staff. The scheduling of hospital personnel is particularly challenging because of different staffing needs on different days and shifts. Because of time-consuming manual scheduling NSP has attracted much attention. NURSE SCHEDULING PROBLEM (NSP) The NSP involves producing a periodic duty roster for nursing staff, subject to a variety of constraints. z A key feature of real NSP is that the planned nurse schedule usually has to be changed to deal with unforeseen circumstances such as staff sickness and emergencies. z NURSE SCHEDULING PROBLEM (NSP) Solution approaches that have been proposed to solve NSP is classified in three main categories: z Optimization-mathematical programming – – z z single-objective (Warner (1976), Rosenbloom, E.S. andGoertzen, (1987)) multi-objective (Arthur and Ravidran,1981) Heuristics (Cheang et al., 2003). Artificial intelligence NSP FORMULATION Indices: g = 1, 2, 3 (nurse grade index) k = 1, 2… N1 (nurse index for grade 1 nurse) l = 1, 2… N2 (nurse index for grade 2 nurse) m = 1, 2… N3 (nurse index for grade 3 nurse) j = 1, 2…21 (shift index) j = 1, 4, 7, 10, 13, 16, 19 corresponds to 7am-3pm shifts j = 2, 5, 8, 11, 14, 17, 20 corresponds to 3pm-11pm shifts j = 3, 6, 9, 12, 15, 18, 21 corresponds to 11pm-7am shifts NSP FORMULATION First Stage Parameters: N1: number of first grade nurses N2: number of second grade nurses N3: number of third grade nurses P1kj: preference cost of first grade nurse k working at shift j P2lj: preference cost of second grade nurse l working at shift j P3mj: preference cost of third grade nurse m working at shift j ADjg: average demand for g th grade nurses at shift j NSP FORMULATION First Stage Decision Variable: ⎧1 if the first grade nurse k is assigned to shift j as a grade g nurse xkjg = ⎨ o/w ⎩0 yljg ⎧1 if the second grade nurse l is assigned to shift j as a grade g nurse =⎨ o/w ⎩0 ⎧1 z mj = ⎨ ⎩0 if the third grade nurse i is assigned o/w to shift j ~ D jg NSP FORMULATION Second Stage Parameters: ocg: over time cost for nurse grade type g ucg: unsatisfied demand cost for nurse grade type g Djg: demand realization for nurse grade type g NSP FORMULATION Second Stage Decision Variable: ox kjg ⎧1 = ⎨ ⎩0 if the first grade nurse i is assigned to shift j as a grade g nurse o/w ⎧1 if the second grade nurse i is assigned to shift j as a grade g nurse oyljg = ⎨ o/w ⎩0 oz mj ⎧1 = ⎨ ⎩0 if the third grade nurse i is assigned o/w to shift ujg: unsatisfied demand for nurse grade type g at shift j j NSP FORMULATION N1 21 3 ∑∑∑ Min k =1 j =1 g =1 P 1 kj x kjg + N2 21 l =1 j =1 g = 2 3 ∑∑∑ P 2 lj y ljg + N3 21 ∑∑ m =1 j =1 ~ P 3 mj z mj + E ( f ( x , y , z , D ) ) Subject to: N1 ∑x k =1 kj1 ≥ AD j1 N1 N2 ∑x k =1 kj 2 + ∑ ylj 2 ≥ ADj 2 N1 ∑x k =1 21 ∀ j = 1,2...21 kj 3 l =1 N2 N3 l =1 m =1 + ∑ y lj 3 + ∑ z mj ≥ AD j 3 3 ∑∑ x j =1 g =1 ∀ j = 1,2...21 kjg =6 ∀ j = 1, 2 ... 21 ∀ k = 1,2...N1 NSP FORMULATION 21 3 ∑∑ j =1 g = 2 21 ∑ j =1 y ljg = 6 ∀ l = 1 , 2 ... N z mj = 6 ∀ m = 1 , 2 ... N xkj1 + xk( j+1)1 + xk( j+2)1 + xkj2 + xk( j+1)2 + xk( j+2)2 + xkj3 + xk( j+1)3 + xk( j+2)3 ≤1 2 3 ∀ k =1,2...N1 j =1, 4,7,10,13,16,19 ylj2 + yl( j+1)2 + yl( j+2)2 + ylj3 + yl( j+1)3 + yl( j+2)3 ≤1 z mj + z m ( j +1) + z m ( j + 2) ≤ 1 x kjg ∈ {0,1} ∀ k = 1,2...N 1 ∀ m = 1,2...N 2 ∀ j = 1,2,...21 g = 1,2,3 z mj ∈ {0,1} ∀ m = 1,2...N 3 ∀ j = 1,2,...21 ∀l =1,2...N2 j =1, 4, 7,10,13,16,19 j = 1, 4, 7, 10, 13, 16, 19 y mjg ∈ {0,1} ∀ m = 1,2...N 2 ∀ j = 1,2,...21 g = 2,3 NSP FORMULATION N 3 21 N1 21 3 N 2 21 3 21 3 ~ E ( f ( x, D ) = Min ∑ ∑ ∑ oc1 ox kjg + ∑ ∑ ∑ oc 2 oy ljg + ∑ ∑ oc 3 oz mj + ∑ ∑ uc g u jg k =1 j =1 g =1 l =1 j =1 g = 2 m =1 j =1 j =1 g =1 Subject to: N1 ∑x k =1 N1 kj1 N1 ∑x k =1 kj 2 N1 ∑x k =1 kj 3 + ∑ oxkj1 + u j1 ≥ D j1 ∀ j = 1,2...21 k =1 N2 N1 N2 l =1 k =1 l =1 + ∑ y lj 2 + ∑ ox kj 2 + ∑ oylj 2 + u j 2 ≥ D j 2 N2 N3 N1 N2 N3 l =1 m =1 k =1 l =1 m =1 + ∑ y lj 3 + ∑ z mj + ∑ ox kj 3 + ∑ oy lj 3 + ∑ oz mj + u j 3 ≥ D j 3 ∀ j = 1,2...21 ∀ j = 1,2...21 NSP FORMULATION ox kj1 + ox kj 2 + ox kj 3 − x k ( j −1)1 − x k ( j −1) 2 − x k ( j −1) 3 ≤ 0 oy lj 2 + oy lj 3 − y l ( j −1) 2 + y l ( j −1) 3 ≤ 0 oz mj 21 − z m ( j −1) ≤ 0 3 ∑∑ ox j =1 g =1 21 3 ∑∑ oy j =1 g = 2 21 ∑ oz j =1 mj ljg ≤1 ≤1 ox kjg ∈ {0,1} ∀ k = 1,2...N 1 oz mj ∈ {0,1} ∀ m = 1,2...N 3 ∀ l = 1,2...N 2 ∀ m = 1 , 2 ... N ≤1 kjg ∀ k = 1,2...N 1 3 ∀ j = 2, 3...21 ∀ j = 2, 3...21 ∀ j = 2 , 3 ... 21 ∀ k = 1,2...N1 ∀ l = 1,2...N 2 ∀ m = 1,2...N 3 ∀ j = 1,2,...21 g = 1,2,3 ∀ j = 1,2,...21 oy mjg ∈ {0,1} ∀ m = 1,2...N 2 ∀ j = 1,2,...21 g = 2,3 u jg ≥ 0, int ∀ j = 1, 2...21 g = 1,2,3 COMPUTATIONAL STUDY z z L2 Algorithm Data Set – – – Number of Nurses (20-25-30) Number of Scenarios Number of Grade one Nurses z Computation Time z Solve the problem using CPLEX CONCLUSION and FUTURE STUDY z z z Improve cut in L2 Algorithm Change demand type Add new constraints THANKS….