How Likely are Your Patients to Return? Unraveling Patient Retention with AI Insights

by | Feb 7, 2024 | Data Insights and Industry Trends

Co-authored by Rosh Raval, GPN Technologies

In the realm of healthcare, just as the costs of welcoming a new patient into your practice are higher than maintaining relationships with established ones, the importance of retaining current patients cannot be overstated. These loyal patients not only cost less to serve but also serve as invaluable ambassadors through word-of-mouth referrals. Losing established patients, often termed as “lost to follow-up” in healthcare or churn in other industries, can significantly impact revenue projections and business strategies. Understanding the factors contributing to patient attrition, whether it’s accessibility issues, dissatisfaction, pricing concerns, or other unmet expectations, offers critical insights for practice improvement.

But how can we anticipate which patients are likely to stay and which may drift away? Enter artificial intelligence (AI) to the rescue! Leveraging AI techniques like probabilistic modeling enables us to predict patient behavior more accurately. Using algorithms such as the Beta Geometric Negative Binomial Distribution (BG-NBD)1 model, we can segment patients based on their likelihood of returning for future visits.

In a recent article,2 we delved into the intricate interplay of visit frequency, recency, and monetary value in shaping patient loyalty. Today, we augment this understanding by integrating the factor of tenure—the duration since a patient’s inaugural visit—into our analytical framework. This holistic approach enriches our predictive capabilities, enabling us to forecast the probability of patient retention with greater accuracy.

This approach acknowledges the reality that patients may disengage from your practice at any time, often without warning. The probabilistic nature of this model embraces uncertainty, offering nuanced predictions tailored to individual patients while remaining applicable to the broader patient population. It’s a departure from traditional deterministic models, providing a more comprehensive understanding of patient behavior and guiding informed decision-making in patient acquisition and retention strategies.


A representative sample of GPN transactions from 2,250 ECPs was used to track 1 million unique patients during multiple visits between 2014 and 2022. Frequencies (counting of repeat transactions) were grouped into 3 bins (“<=3”, “4-6” and “7+”). Both recency (years since last visit) and tenure (years since first visit) are also banded into 3 groups (“<=3 Years”, “4-6 Years” and “7+ Years”). The monetary value (median list dollar revenue for all transactions by the same patient over the time frame) is condensed as (“<$1K”, “$1K-$2.49K”, or “$2.5K+”).


The outcome here is the chance (i.e., likelihood, or probability) of each of the 1 million patients for re-visiting their optometrist. While we are measuring a binary outcome, “will or will not visit again”, the probability of this outcome can range from 0 to 1. In this sample, the average probability was 0.56 while the median probability was 0.67. 

RE-VISIT LIKELIHOODAverageMedianMinimumMaximumRange
Unlikely0. – 0.24
Maybe0.370.370.250.490.25 – 0.49
Likely0.810.850.500.990.50 – 1.00
Overall Average0.560.67nanana
Table 1. Statistics of re-visit likelihood bins


Correlations are measures of the directionality between values and vary from -1 (strong inverse correlation) to +1 (strong positive correlation) with zero meaning no correlation. 

Recency has the strongest correlation to re-visit likelihood: 0.66, a relationship that says “the longer since the most recent visit, the more likely to re-visit”.  Frequency and Monetary value (0.43 and 0.46 respectively) are also strong drivers of the probability to re-visit.  Tenure (-0.10) correlation is weak and inverse, meaning and probability of continued visits is very slightly aligned with more recent visits.

Key DriverCorrelation
Table 2. Correlations of Key Drivers to re-visit probability.


Looking at the distribution of patients by re-visit probability (area chart on the left) we see a bimodal distribution with two peaks, one around the more than 5,000 patients with about a 5% chance of re-visiting.  Another peak begins around the higher probability values (0.7 and greater). The distribution is not symmetric; it is skewed to the right (more folks likely to return) and the higher probability values are more spread out.


Grouping the same probability to re-visit values into categories labeled “Unlikely” (probability <0.25), “Maybe” (probability 0.25 – 0.49) or “Likely” (probability 0.5 to 1) we see that the majority of patients in this sample (61%) are likely to return while 27% are unlikely to return and 12% may return.


The independent variables of Recency, Frequency, Monetary and Tenure are displayed as slicers above both visualizations. Choosing any of three levels of any of these variables impacts the likelihood outcomes shown in the 2 chart visualizations:

    • 7+ Years: When we filter the results to those patients whose last visit was 7 or more years ago, the vast majority of patients are likely to return (96.8%), with just 2.1% classified as maybe and 1% as unlikely.
    • 4-6 Years:  Patients with last visit 4-6 Years ago are also skewed toward more likely to re-visit, likely to re-visit comprises 72% of the sample; 17% of patients with 4-6 years since visiting are “maybes”, and 11.1% are unlikely to re-visit.
    • <= 3 Years: Looking solely at patients seen within the prior 3 years tenure or less, the peaks are all shifted to the less likely to re-visit side of the area chart with the column chart showing 44.1% of these patients unlikely to return while 15.5% fall into the maybe category and 40.3% are likely to return.
    • 7+ Years: The bi-modal distribution shown in the left area chart visual become more polarized when we filter the results to those patients who began visiting 7 or more years ago. And, the column chart on the right, when filtered to 7+ years ago shows 55.7% are likely to come back, while 10.1% may return and 34.2% are unlikely to repeat visits.
    • 4-6 Years:  Patients with 4-6 Years Tenure also show a bipolar distribution,  with the strong likelihood of repeat visits accounting for 62.2% patients; 12.6% of patients with 4-6 Years tenure are “maybes”, and 25.2% are unlikely to re-visit.
    • <= 3 Years: When we narrow the results to only those with 3 years tenure or less, the peaks are all shifted to the more likely to re-visit side of the area chart with the column chart showing 72.6% of patients with <= 3 Years Tenure are likely to return while 17.3% fall into the maybe category and 10.1 are % unlikely to return.
    • 7+ Visits Frequency: Patients with a history of 7+ visits exhibit a high 86.8% likelihood of repeat visits, 3.8% are a maybe, and 9.4% are unlikely to repeat visits.
    • 4-6 Visits Frequency: 76.3% of patients with 4-6 prior visits are likely to return, while 15.4% are unlikely and 8.3% are maybes.
    • <= 3 Visits Frequency: 46.7% of patients with a frequency of 3 or fewer visits are likely to return, 36.6% unlikely and 16.7%  are considered maybes.
    • $2.5K+ Spending: 89.4% of patients spending $2.5K+ over their lifetime are likely to return. Only 6.6% of patients spending this much are unlikely to return and 3.9% are maybes.
    • $1K-$2.49K Spending: A large share (70.9%) of those who have spent between $1K-$2.49K also show a strong likelihood of repeat visits; 18.3% of this spend level are unlikely to return and 10.8% are a maybe.
    • <$1K Spending: Among patients spending less than $1K there is an even split: 41.7% have a likelihood of repeat visits while 41.5% are unlikely, with 16.9% classified as maybe.

Exploring the interplay of key drivers influencing patient re-visits reveals a multifaceted picture. By combining variables like recency, tenure, frequency, and monetary value, we can ascertain the likelihood of patients returning. Visualizing the data through slicers unveils intriguing insights—for instance, when recency and tenure span 4-6 years, with a frequency of 4-6 visits and a monetary value exceeding $2.5K, we observe the highest probability of re-visitation. With 81 unique combinations at our disposal, each offering distinct probabilities and shares of returning patients, the potential for strategic refinement is vast.

Delving deeper, it’s intuitive that those inclined to return tend to have been absent recently, exhibit frequent past visits, and often spend more. Segmenting patients based on re-visit likelihood aids in identifying those at risk of disengagement—those who haven’t visited recently, visit infrequently, or spend less. Deploying proactive campaigns with personalized messaging and tailored marketing can bolster retention efforts significantly.

Moreover, here are additional strategies to amplify the chances of patient re-visits:

  1. Enhance Accessibility: Streamline appointment scheduling, minimize wait times, and ensure ease of access to your office, as patient convenience is paramount.
  2. Communication is Key: Stay active on social media platforms, keeping patients informed about office updates, staff changes, and new services.
  3. Implement Loyalty Programs: Encourage repeat visits by offering incentives to loyal patrons.
  4. Elevate Patient Value: Provide value-added services tailored to individual needs, without resorting to pressure tactics.
  5. Monitor Feedback: Actively address patient feedback and reviews, leveraging constructive criticism to refine your services.

By adhering to these principles and monitoring predictive indicators, you can optimize the likelihood of patient re-visits, fostering enduring relationships and sustaining a thriving practice.

  1. ↩︎
  2. ↩︎

About Industry Trends

Through robust analysis of anonymized data, we are able to develop insights, profiles, and a deeper understanding of market results and benchmarks.

GPN aggregates millions of transactions from thousands of eyecare providers, and focalCenter performs rigorous analysis for delivering timely and precise micro and macro dashboards with interactive business intelligence to the eyecare industry. Please feel free to contact us for more information on growing your eyecare business with data-driven strategies.

By Ron Krefman, OD

Finding solutions in data science.


Share This