Article Text

Download PDFPDF

22 From no show to arrived: using machine learning to bolster patient attendance for resident continuity-clinic appointments
Free
  1. Jaishree Hariharan1,
  2. Roma Bhatia2,
  3. Zachary Lenhart1,
  4. Janice Cunningham1,
  5. Brea Sodini1,
  6. Dang Tran1,
  7. Oscar Marrequin1,
  8. Gary Fischer1
  1. 1University of Pittsburgh Medical Center, USA
  2. 2Beth Israel Deaconess Medical Center, USA

Abstract

Background Resident continuity-clinic (RCC) is a crucial component of ambulatory training in primary care.The no-show rate (NSR) in a large academic center with 60 residents averaged 27% in academic year (AY) 2018, despite an automated phone/text reminder system 3 days prior to appointment, resulting in fragmented care, reduced access and decreased learning opportunities for residents.

Objectives To determine whether telephone outreach targeting patients predicted to be at high-risk to no-show can reduce NSR for RCC appointments.

Methods A validated machine-learning prediction model developed by data scientists at UPMC for Primary care, generated a daily list of high-risk patients (i.e. =20% risk to no-show). Starting Oct 2018, these patients received a phone-call reminder from a clinical staff, 48 hours prior to their scheduled appointment. The outcomes of the call recorded were confirmed, cancelled, rescheduled, voicemail, not reached. Monthly NSR was tracked from July 2017 through June 2019 and analyzed using control charts.

Results Fifty-nine percent (1206/2046) of targeted patients were reached. Of those 89% confirmed and 10% canceled or rescheduled their appointment. The overall no-show rate for RCC appointments in (AY) 2019 decreased to 23%, p<0.01,95% CI [21.6% to 25.0%], resulting in additional 283 completed visits and $40,000 in revenue. Higher no-show rates correlated with lower percentage of patients reached. Patients on government-assisted insurance (76%) and African-Americans (61%) had higher no-shows and a major barrier was transportation.

Abstract 22 Figure 1

Determinants of machine learning algorithm

Abstract 22 Figure 2

Sample daily report

Abstract 22 Figure 3

Outcome of Phone calls: Reached or Not reached. If reached, appointment confirmed, canceled, rescheduled, or other (language barriers/hospital admissions). If not reached, other includes: no answer/ no voice mail set up, phone number wrong in chart, & no active phone

Abstract 22 Figure 4

Analysis of Resident clinic Appointments: AY 2018 and AY 2018. No-show rate (NSR) calculated against completed + no-shows. * P value using Chi-Square test. ** Canceled visits include same day cancellation and cancelled rate calculated against ‘All’ visits. ±±±

Abstract 22 Figure 5

(A) Control Chart (P chart): Resident Clinic No-Show data from July 2017–June 2019. (B) Control Chart Showing: No-show rate, pre (AY 2018) and post intervention (AY 2019). Intervention start date -Oct 2018

Abstract 22 Figure 6

Run Chart (Oct 18–June 19): Model Predicted No Shows (N), patients reached (N&%), monthly NSR%. Apr->Jun; lower percentage of patients reached and higher no-show rates

Abstract 22 Table 1

Patient characteristics: resident clinic. Unique patients scheduled in AY 2019

Conclusions To our knowledge this is the first study showing that targeted phone outreach for high-risk patients can decrease NSR for RCC appointments, augmenting resident learning opportunities and revenue.

Statistics from Altmetric.com

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.