Healthcare Marketing – AI & Patient Engagement Nearly Doubling Email Engagement

Healthcare Marketing: Natural Language Processing (NLP) analysis uses 2M+ emails to discover how to beat patient engagement benchmarks

At SymphonyRM, we obsess over how to use data science for Health Systems and Healthcare Marketing to find Next Best Actions (NBAs) for patients. Once we identify NBAs, we focus on how to use text messaging, email, the access center, or many other channels to engage patients. Email is an excellent, cost-effective channel to reach large groups of patients. Using Natural Language Processing (NLP) on 2 million+ newsletter emails, we share how to help your outreach exceed industry benchmarks.

But first…

What is a Next Best Action?

Next Best Actions are simply actions patients should take given their patient journey and risk factors. NBAs can take on many forms: from reading a newsletter item or attending a seminar to getting a screening or scheduling an appointment. We assign NBAs using EMR & engagement data, service line capacity, and the contracts and goals of the health system.

Even without the data science and structure to find and measure NBAs, every health system targets actions specific to each patient that will help them in their care. Think of Next Best Actions as a data-driven way to define and track these actions. We use Natural Language Processing (NLP) and other data science techniques to refine engagement approaches that drive adoption.

Refining email campaign outreach

To refine this approach for email newsletter campaigns, we looked at a bevy of characteristics. These include words in the subject line, date of launch, type of email, sender of email, content & illnesses discussed, and subject line length. Then, we look at how these characteristics impact click through rates,  open rates, and conversions. We collaborate with Healthcare Marketing and other teams to execute these campaigns.

That sounds like a major effort, is it worth it?

In the healthcare industry, email open rates average 21.72%, and click-through rates average 2.49%.¹ Campaigns with SymphonyRM’s refinements perform at 30.33% open rates and 3.93% click-through. This represents performance 40% to 55% higher than industry standards!

…people who receive & interact with these email campaigns book appointments at a 5% higher rate than those who don’t.

More importantly, people who receive & interact with these email campaigns book appointments at a 5% higher rate than those who don’t. This means the right messaging can convert a new appointment emailing only 20 patients. This is critical to your goals to close care gaps, reduce costs, and drive necessary procedures in your community.

So, to answer the question above, yes, this level of constant monitoring and analysis is vital.  It helps drive the right messaging to patients and stay aligned with their needs. To break down what it takes to exceed patient engagement expectations, we’ll start with where we found the biggest impact.

Who sent you?

Of all these traits, one stood out the most: the sender. In particular, whether the message is from the provider or the organization:

One test we use for campaigns is randomly choosing the clinic or a patient’s Primary Care Provider as sender. The above chart shows that patients have a clear preference for messages that come from their providers.

This is a powerful tool, but it’s still important maintain a balance. For high risk, high priority Next Best Actions, it’s worthwhile to send as provider. However, lower priority communications can take on other names to prevent message fatigue.²

While adjusting the “Sender” offers the biggest impact, other factors are important. Next, we focus on what makes an effective subject line.

Subject Lines: Choosing Your Words

It turns out that some words are more impactful than others at driving engagement.

The word cloud above ranks each word by size and its association with high performing campaigns.

The top performing word is the patient’s first name, and it’s associated with 3.7% higher click-thru rates. All five top performing words (highlighted in blue) performed with at least 2.6% lift. They’re telling Healthcare Marketing to focus on health and wellness news and to personalize to the patient’s name and the time of year.

The bottom performing 15 words are highlighted in orange and have the smallest font, and mid-tier performers appear in grey. These results show that using high engagement terms is key to closing as many Next Best Actions as possible.

Subject Lines: Finding the Ideal Length

…the longer the subject line, the less likely patients were to click.

Past about 35 characters, the longer the subject line,  the less likely patients were to click.³ People tend to first see emails as notifications on mobile devices. This means a compelling message in a small space helps patients digest and engage.

Controlling for other factors that impact engagement rates, such as the message sender, we found that a subject line should be between 25 and 40 characters.

Health Issues to Discuss

Each newsletter allows you to discuss a wide range of issues with your audience. Typically, there’s a central topical focus as well as additional topics to cover other matters impacting community health.

From the study, here are some of the keywords that were associated with higher performing campaigns: infection, cancer deaths, diabetic retinopathy, cancers, heartburn, opiate pain, anxiety, abuse, digestive disorders, prostate cancer, weight loss, suicide, colorectal cancer, and sleep disorders.

…issues people are likely to engage reflect important healthcare interests and needs in a given community.

This isn’t to say that any of these should be considered for use in engagement campaigns. Rather, we’re saying issues people are likely to engage in reflect important healthcare interests and needs in a given community. Additionally, engagement at this level can help inform what Next Best Actions may be appropriate for patients.

Consider your timing

Seasonality is definitely a factor concerning illnesses & admissions⁴, and it follows that these would correspond engagement rates. Here’s the seasonal breakdown for the health systems in our study:

Your health system won’t break down exactly like this, but it’s important to understand when your community is most responsive to which communications. This is especially for diseases that have strong seasonality (e.g. the flu).

This helps healthcare marketing to plan and structure campaigns based on when they can deliver the most impact for patients and value to the health system.

What about OUR Data?

We’ve presented immediately usable, high-level steps to help your health system’s campaigns drive more action among patients.

The best way to execute the Next Best Actions you’re already targeting in your community is from your own data.

We derived these insights from multiple health systems, but behaviors and preferences change over time. The best way to execute the Next Best Actions you’re already targeting in your community is from your own data.

Feel free to use the insights from this article. Also, if you’re interested in what Next Best Actions and optimizations may be hiding in your own data, let us know here or by emailing info@symphonyrm.com.

Major thanks to Sarah Coles, a visionary in data-driven patient engagement and provided guidance for this work.

References:

1 – Email Engagement Rates by Industry, Mailchimp

2 – APA PsycNET, So, J., Kim, S., & Cohen, H. (2017)

3 – Our minimum subject line length was 16 characters, and the maximum was 74. This means that further experimentation could show advantages with 1- or 2-word subject lines, as these weren’t present in this study

4 – Trend and Seasonality in Hospitalizations for Pulmonary Embolism: A Time Series Analysis – R. Guijarro (2014)

Specialized Open Source Libraries:

pandas – Analysis & preprocessing

matplotlib – Plotting

seaborn – Statistical plots

WordCloud – Word cloud visualization

spaCy – Natural Language Processing – Preprocessing

ScispaCy – Biomedical Natural Language Processing

scikit-learn – Machine learning models & NLP vectorization

StatsModels – Statistical analysis

numpy – Data processing & matrix math

DoWhy – Causal inference