Predictive Analytics for Capacity Planning HIC 2015 Andrae Gaeth - - PowerPoint PPT Presentation

Evaluating Predictive Analytics for Capacity Planning

HIC 2015 Andrae Gaeth

Predictive analytics is the practice of extracting information from existing data sets, and then applying various techniques (eg, statistical, modelling) in order to determine patterns and predict future outcomes and trends. How can we practically evaluate and use predictive analytics solutions for capacity planning within health?

What is predictive analytics?

Paradigm Shift Required for Analytics Maturity

Sources: Competing on Analytics, Tom Davenport Gartner IT Glossary

What happened? What will happen? What should I do? Why did it happen? What is happening? Descriptive Predictive Prescriptive Real Time Descriptive

What to look for with Capacity Planning

• 1. Manages multiple planning horizons

– Multi-year, annual, 6-8 week scheduling periods, weeks, days and hours

• 2. Continuously updates forecasts
• 3. Forecasts patient demand with consistently high

absolute accuracy vs. stated accuracy % aggregated over time

• 4. Forecasts volumes for door-to-door patient flow
• vs. preset intervals and departments
• 5. Converts forecasts automatically into capacity

and staffing needs

• 6. Supports user insight, input and adjustment as

part of the planning process

Strategic & Annual Planning

• Discuss “what-if” options, regional plan
• Budget and physical capacity decisions
• Set targets and assumption (linking plans)

Monthly & Weekly Planning

• Manage “current” variation to plan
• Update forecasts & roster staff
• Informed decision making

Daily Planning

• Unit focus: manage current & projected pts.
• Focus on relieving immediate flow issues
• Replace sick leave? Book casual

Forecasting Methods Used for Operational Planning

Predictive modeling Algorithmic modeling Pattern identification Scenario modeling Simulation Optimization

Predictive

Determining, mathematically, the relationship between the explanatory variables and the predicted variable, based

• n historical data

Examples: Insurance: relationship between certain characteristics of a person and lifestyle and predicted outcome of a certain claim Capacity Planning: relationship between status of ED at a given time of day and the impact on Inpatient beds tomorrow

Pattern Identification

‘Time series forecasting’ identifies different patterns within the predicted variable itself, such as trend, seasonality and day of the week. Various combinations of those patterns can then be used to derive a forecast

Algorithmic

Take recent averages and distributions of past activity for certain locations, services and day of the week or special event days. Then apply an algorithm to utilize these to create a projection

2012 2013 2014 2013 2012 2013 2014 145,201 150,721 151,870 28,384 5.6 5.3 5.0 122,548 130,929 133,984 23,016 6.1 5.7 5.3 22,653 19,792 17,887 5,368 4.1 3.7 3.5 4,229 4,984 5,546 764 7.6 6.5 7.0 4,113 4,509 5,304 727 7.6 6.2 7.0 115 475 242 37 6.8 12.8 7.8 2,732 2,764 2,536 1,156 2.4 2.4 2.2 1,510 1,630 1,506 615 3.0 2.6 2.5 1,222 1,134 1,030 541 2.0 2.1 1.8 14,197 14,147 14,058 3,488 4.3 4.1 4.0 9,128 10,111 10,041 2,314 4.6 4.4 4.3 5,068 4,036 4,017 1,174 3.9 3.4 3.3 43,161 43,830 41,924 6,800 6.5 6.4 6.0 31,140 33,538 32,961 4,549 7.3 7.4 6.7 12,021 10,292 8,963 2,251 5.0 4.6 4.4 Emergency 4,265 4,884 Scheduled 2,410 2,059 IP Surgery 6,675 6,943 Emergency 1,968 2,354 Scheduled 1,302 1,200 IP Cardiac 3,270 3,554 Emergency 543 756 Scheduled 17 31 ICU 560 787 Emergency 511 606 Scheduled 614 564 Cardiac ICU 1,125 1,170 25,266 Scheduled 5,555 5,042 Annual Patient Turn 2012 2014 Hospital Total 25,756 30,308 Emergency 20,201 Annual Census Annual Patients In

Scenario

Bed days 4% Activity 16% ALOS 12%

• Analyze which services were driving these trends?
• What do we expect this year? Sustain or continue trends?
• What change initiatives are we investing in to increase volumes and

reduce ALOS?

Allow users to input their own assumptions or ‘what ifs’ into the prediction to assess impacts (vs. based purely on patterns of past activity)

Simulation

Coded logic which emulates the actual individual patient ADT activity of planned patient admissions/bookings

Using either a mathematical or algorithmic approach to derive the optimal

• utcome, based on an ‘objective function’.

‘Business as usual’ would have required 50 – 60 beds staffed; using an optimised activity plan resulted in 40 - 50

04/04/2013 04/08/2013 04/12/2013 04/16/2013 04/20/2013 04/24/2013 04/28/2013 05/02/2013 05/06/2013 05/10/2013 05/14/2013 05/18/2013 05/22/2013 05/26/2013 05/30/2013 06/03/2013 06/07/2013 06/11/2013 06/15/2013 06/19/2013 06/23/2013

Optimisation

Patient pathways leveraged to simulate and/or optimise any impact on OR’s and inpatient units.

Pathways

Develops base patient demand based on current trends seen from historical data.

Historical Data

Algorithms applied to forecast increasing accuracy of patient demand.

Algorithms

Combining Approaches

Customers clinical expert input of areas that may impact any patient demand added.

Clinical Input

Aggregate vs. Patient-level Static vs. Dynamic Short term vs multi-horizon Top down vs. Bottom up

Comparing Planning Approaches