Needs-based population segmentation David Matchar, MD, FACP, FAMS Duke Medicine (General Internal Medicine) Duke-NUS Medical School (Health Services and Systems Research)
Service mismatch: Many beds filled with people not needing acute care Based on reviewing case notes of Geriatrics ward in an acute hospital in Singapore 2
The quadruple aim Population health Positive patient experience Low per capita costs Satisfying provider work life Needs-based segmentation provides a foundation for dynamically rebalancing health services to achieve the quadruple aim Optimal population health is achieved by meeting healthcare needs (Bodenheimer et al, 2014)
What is a healthcare need? An ability to benefit from a healthcare service A reduction in probability of progression from healthy to increasingly worse need states It is a prediction, not an absolute certainty In an imperfect market, it is often distinct from supply and demand 4
Healthcare services to fulfill needs 1. Non-skilled home services 2. Social support 3. Physiological monitoring & prompt follow-up 4. Regular physician services 5. Medication management 6. Supervisory care 7. Nursing type skilled services 8. Rehabilitation type skilled services 9. Care coordination 10. Patient skills education 11. Caregiver skills education 12. Palliative care What features of an individual identifies which services are needed? 5
1. Global health features related to needs 2. Complicating features related to needs Healthy Chronic condition but normal function Frailty with or without dementia 1. ADL/IADL deficits 2. Nursing-type task needs 3. Rehabilitation-type Stable with task needs serious 4. Organization of care disability 5. Activation in own care Limited reserve 6. Disruptive behavioral/psychiatric and issues exacerbations 7. Social support in case of need 8. Hospital admissions in past 6 months 9. Polypharmacy Dead Short period of decline before dying 6
How? Needs-based population segmentation 7
Proposed tool requirements Simple and fast Strong face validity Good inter-rater reliability Good predictive validity Parsimonious Linked to action 8
Identified adult population healthcare need based segmentation tools S/N Tool Number of articles 1 Adjusted Clinical Groups 17 2 Clinical Risk Groups 6 3 Senior Segmentation Algorithm 1 4 Bridges to Health 1 5 LCA1 - Demand driven care 1 6 LCA2 SIPA Trial 2 7 LCA3 - Taiwan National Health Insurance 2 8 Complexedex 3 9 Lombardy Segmentation 1 10 North West London Segmentation 1 11 Delaware Segmentation 1 Source Systematic Review Systematic Review Systematic Review Systematic Review Systematic Review Systematic Review Hand searching Hand searching Hand searching Google searching Google searching
Comparison of segmentation tool segment concepts
Needs-based population segmentation seems like a fairly straightforward task However, Most segmentation schemes are aimed at identifying utilization and risk, not need Most are based on use by people who show up for services Virtually all require an EMR which does not generally provide basic health-related social features None satisfied the proposed tool requirements 11
Strategies for population segmentation Segment by utilization Segment by need High Use Not High Use No unmet needs Unmet needs 12
Elements of the SST 6 global health features Based on the Bridges to Health 9 (actionable) complicating features 1. ADL/IADL deficits 2. Nursing-type task needs 3. Rehabilitation-type task needs 4. Organization of care 5. Activation in own care 6. Disruptive behavioral/psychiatric issues 7. Social support in case of need 8. Hospital admissions in past 6 months 9. Polypharmacy 13
Predictive validity: time to hospital readmission
Predictive validity: time to ED visit
SST predictive validity in the inpatient setting SST variable Cox regression for time to emergency department visit or death in 90 days controlled for age and gender (N = 216) Cox regression for time to non-elective hospital admission or death in 90 days, controlled for age and gender (N = 216) Hazard Ratio between highest and lowest levels Hazard Ratio between highest and lowest levels Global impression of patient 5.201** 5.398** Acutely ill but curable condition 1.344 1.106 Functional assessment 1.882* 1.273 Skilled nursing type task needs 2.118** 1.732* Organization of care 4.954*** 3.534*** Activation in own care 3.029*** 2.258*** Disruptive behavioral issues 4.708*** 3.096*** Social support in case of need 1.481 1.369 Hospital admissions in last 6 months 2.570*** 2.214** Polypharmacy 1.990* 2.382* Complexity count categories 4.688*** 2.870*** * p < 0.05, ** p < 0.01, *** p < 0.001
Community survey version of SST: outcomes in 180 days among low-income elderly Hospital admissions by SST Segments Mortality by SST Segments 0.2.4.6.8 1 Probability of mortality in 180 days 0.1.2.3 Healthy Asymptomatic Symptomatic Decline Limited Reserve SST Global Impression Healthy Asymptomatic Symptomatic Decline Limited Reserve SST Global Impression (mean) Admissions2017 yu/yl (mean) any_mort yu/yl
But how do we use the SST to identify unmet needs for health and health-related social services? 1. Non-skilled home services 2. Social support 3. Physiological monitoring & prompt follow-up 4. Regular physician services 5. Medication management 6. Supervisory care 7. Nursing type skilled services 8. Rehabilitation type skilled services 9. Care coordination 10. Patient skills education 11. Caregiver skills education 12. Palliative care 18
Definition of service functions & primary indications (examples) Function Definition Primary Indicators Non-skilled home Home care to meet basic and instrumental activities of daily services living (e.g., assistance with dressing, bathing, meal preparation) Any IADL or ADL Social support Care coordination Physiological monitoring & prompt followup Any support that aids patients with decision making, companionship/empowerment, or basic healthcare services (e.g., befriending services) Systematic interfacing between multiple providers (e.g., patient navigator). Frequent (i.e., daily to weekly) monitoring of physiological signs (e.g., weight, dyspnea, blood pressure) and prescribed responses to abnormalities (e.g., diuretic adjustments for CHF, rescue medications for COPD) or first level response leading to definitive prescribed response. Low social support, level 1 and 2 Organization of care, level 2 GI: limited reserve...
20
21
4 2 3 1 1 1 2 1 3 3 Primary indications Prototypical service needs Secondary indications Appropriate means 22
SST algorithm Program
Conclusions The Simple Segmentation Tool is designed to be used to segment patient population based on their health and social service need characteristics. The instrument was developed as no other similar tool can be found in the literature. The SST has been validated in terms of inter-rater reliability, convergent validity and predictive validity when used by clinicians such as doctors and nurses. The SST can also be used to segment populations using community based survey mapped data elements The acid test of segmentation is whether it is predictive of response to interventions
Future Work Track the epidemiology of needs over time Assess the impact of concordance with norms SST assessed for 1,000 patients discharged from medical ward Identify service use at 3 months vs. norms Assess whether concordance between use and norms is predictive of outcomes at 1 year Translate the SST to the Durham context Link to big data 25
Thank You 26
Comparing SST with other schemes Hospital Admissions by Segmentation Schemes Mortality by Segmentation Schemes.5 1.5 0 1 2 Probability of Mortality in 180 days 0.1.2.3.4 1 2 3 4 5 6 7 8 91011121314 1 2 3 4 5 6 7 8 910 1 2 3 4 5 6 1 2 3 4 5 6 7 Singapore Delaware Lombardy NW London All Segments y yu/yl 1 2 3 4 5 6 7 8 91011121314 1 2 3 4 5 6 7 8 910 1 2 3 4 5 6 1 2 3 4 5 6 7 Singapore Delaware Lombardy NW London All Segments y yu/yl 27
Number of doctors Frequency of seeing the same doctor Length of consultation Stable chronic condition Service gap Doctorpatient relationship Model of care efficiency Visit waiting time Building relationships Clinic volume IV Provider work life Population health I Population health Complicated chronic condition III Percapita cost Change in positive patient experience II Positive patient experience Attractiveness of clinic Out-of-pocket cost 28