15.

760: University Health Service

1. Admin: Memory Jogger, The Goal, Mike Hammer 2. What are the sources of variability at UHS? 3. What are the problems UHS is experiencing that should be addressed? 4. Describe the Process Flow Diagram. 5. How to usefully model this system? 6. Utilization and waiting for nurse practitioners. 7. Utilization and waiting for physicians. 8. Recommendations

UHS Process Flow Diagram

Patients arrive

Front desk
NP wait

AVF form
Nurse Practioners

Wait for records

Wait for Triage

Triage

Exit

Any MD wait

Dr. A Dr. B Dr. C

sources of variabi l i t y
Patients arrival rates by hour, day, week and season. Number of MD's on duty - by time of day, day of week. Number of NP's on duty - by time of day, day of week. MD service rates by Doctor and patient ailment NP service rate by nurse and patient ailment. Triage service rate. Triage coordinator allocation to MD - NPs.

Others - # of patients wanting to see a special NP or MD, - by hour, day, week. - Speed of filling out AVF forms.

Patient Arrival Rates

The clinic is opened to patients from 8:00AM to 5:30PM, a total of 9.5 hrs. Staff are asked to stay until 6:00 PM, So the clinic serves patients 10 hours per day. For simplification, we will assume that patients arrive over the 10 hrs. Patient average arrival rate = =(143 pat/day)/10 hrs/day = 14.3 pat/hr 8 - 9:00 AM on Monday mornings is a peak hour. Patient peak arrival rate = = 14.3 [average rate] X (163/143)[the Monday factor] X (18.2/14.3)[the 8-9 AM factor] = 20.7 patients/hr

ASSUMPTIONS OF THE QUEUEING MODELS

Poisson arr ivals /exponent ialservice t imes steady state <1 , when com puting t he queue lengths and wait i ng t i m es Constant # of servers FIFO serv i ce Singlel ine queue ( to MD's) In f in i te queue capacity Ignore specia l pr ior i t y emergencies Ignore special priority requests

Tr iage and ar r i valrates

Let si nterpret the second colum n of data i n Exhibi t6 as fo l l ows: 28 % of pat ients are t r iaged to t he NP queue, 48 % are t r iaged to the queue f or f i rst ava i l able M D, and 24 % queue up for a specif ic MD. ( I .e . ,l et s assu m e the f ract ion of pat ients thatrequest to see a speci f ic NP is zero. ) Ho wever ,s i nce 5% of pat ients seen by NPs get sent on to M Ds ( bottom of p.4 ) ,these pat ients wai ttwice and are seen twice, so t he totalload on t he system might be thought of as greater t han 100 % ofin i t ial demand. Therefore,let s assu me thatthe percentage of pat ients who see an M D = 24 % + 48% + [ (28%) x 0.05] = 73.4% o f in i t ia larr iva l s and the arr iva lr ate to the NP queue is 28% ofi ni t ia larr iva l s.

Arrival Rates
Pat ients average arr iva lrate t o an NP Average over week, = 14.3 pat /hr (28%) = 4 . 0 pat/hr . Average over peak hour = 20.7 pat /hr (28%) = 5 . 8 pat/hr . Pat ients average arr iva lrate t o an M D Average over week, = 14.3 pat /hr (73 .4%) = 10.5 pat /hr Av. over peak hour = 20.7 pat/hr (73.4%) = 15.2 pat/hr

Num b er of Servers

(From exhibi t4) W eekly average nu mber ofNP servers = 3 . 15 Average nu m ber of NP servers dur ing peak Monday hour = 2 W eekly average nu mber ofM D servers = 2 . 9 Average number of MD servers during peak Monday hour = 2

ut i l i zat i on rate ,length of queue and waiting t ime for

the M D's and NP's . Sadly, as in real l i fe, data and models are imper fect . As a resul t , one must of t en make decisions with ad mit tedly imperfect understanding of the systems under st udy. The data in the case permit a range of analyt ica lapproaches,t wo of which are lustrated be low. i l The f i rs tm ethod uses as inputs t he expected serv i ce t i m e, pat ient arriva lr ates and nu m ber of servers g iven in t he case to co m pute ut i l i z at ion ra tes. The second m ethod uses an inputs wait ing t ime and pat ients arriva lr ate f rom the case t o com pute the queue length using Li t t le'sform ula. Then, the queue l ength co m bined wi t h the nu m ber of servers gives the ut i l i z at ion ra te using table A2.

M ETH O D #1 M, ,S ( f r o m case) r by def ini t ion, = M/S , S Lf rom Table A2. L, W(*)with Li t t le'sform ula, W = L / ( * com pare wi t h case data) Method #2 W, L wi t h Li t t le 's f ormula ,L= W L, S , f rom Table A2 , , S M = /S by defin i t ion

NP Calculat ions A. Method #1 Average f or the Week. M = 32.8 min /pat ient (Ex5) =.547 hrs/pat S = 3.15 ( Ex4) =4 . 0 patients/hour (above page) = M/S = (4 . 0) ( .547) /3 . 15 = . 695 Fro m tab l e A-2, we have L = 1.15 pat ients W = L/ = 1.15 pat/4.0 pat/hour = .29 hrs = 17 min

Peak Hour M = 32.8 min /pat(above) = .547 hr / pat S = 2 (Exhib i t 4) =5 . 8 patients/hour (above page) = M/S = (5.8)(.547)/2 = 1.58

Method 2: Nurse Pract i t i oners Average f or the Week W = 6.7 min =. 112 hrs L = W = 4.0 pat /hr ( 0.112 hrs)= .418 patients . . S = 3.15 . f rom Table A2, = 0 .57 M = S/ = (3 . 15) (0.57) /4.0 =. 449 hrs /pat Thus the serv i ce rateis 1/ ( .449) = 2.23 pat /hr and /M = .57 / .449 = 1 .27 pat /hr , Peak Hour S = 2. M =. 449 hrs /pat ,s ince we assu me the sa me serv i ce rate as dur ing regular hours,i .e . ,the i r service rate i s independent of the patients ar r ival ra t e. = M/S = (5.8)(.449)/2 = 1.30, again greater than one, so inventory builds up.

Method #1 : MD Calculations Average f or week M = 19.4 minutes/patient = .323 hrs/pat i ent S = 2.9(exhibi t4) = 10.5 pat /hr = M/S = 10 .5( .323) / (2 . 9) = 1.17,implying that queues bui ld al l day. Thisis suspic ious since there is no indicat i on that som e pat ients ca m p overnight or are sent ho me. Perhaps: -t r iage sends pat ients t o NP's when the M D's are busy. - Staf fs tays la t e Peak Hour M =. 323 hrs /pat ient , S =2 = 15.2 pat /hr , = (15.2)(.323)/2 = 2.45 !!

M D CALCULATIONS: Method #2 Average f or Week W = 25.2 min = .42 hrs L = W = 10.5 pat /hr ( .42 hr )=> 4.4 patin queue. Using S = 2 .9 and L = 4 .4 in Table A-2 g i ves = .86 M = Sr/ l = (2 .9 ) ( .86) / 10.5 = .238 hrs/pat Thus the serv i ce rateis 1/ .238 = 4 . 21 pat /hr and /M = ( .86)/ .238 = 3 .61 PATIENTS/hr, nei ther ofwhich is very close to the reported value o f3 . 1 patients per hour .

LEAR NING POINTS 1. Queueing m odels do not predic tthe customer wait i ng t i m es with great accuracy. This could be because one or more ofthe assum ptions under ly ing the model was seriously vio lated,or i tcould be thati naccurate data was reported in the case. 2. Analysis suggests thatt he MD's were m ore heavilyloaded t han t he NP's ,and t hat t he ent i r e staf f was heavi lyl oaded during peak hours. 3. The models provide a structure f or th i nking aboutthe operat ing syste m. The f ormulas L = W , = M, and W = (M 2 + 2) /2 ( 1-) do stimulate thought about the i m portant param eters ofthe Syste m,their r elat i onships wi th each other and t he decis i on var iables avai lable t o Ms. Angel l .