Why is it important to document missed appointments in the patients medical record?

Skip to main content

This browser is no longer supported.

Upgrade to Microsoft Edge to take advantage of the latest features, security updates, and technical support.

Overview and prerequisites for missed appointments (preview)

• Article
• 08/19/2022
• 6 minutes to read

In this article

Important

Some or all of this functionality is available as part of a preview release. The content and the functionality are subject to change.

The missed appointments feature is based on an AI model that helps organizations assess the likelihood that patients will miss their next appointment. The predictions can be used by healthcare providers to take proactive measures that help ensure that patients will attend their next appointment and maintain continuity of care. Missed appointments are costly for clinics, and research has shown that patients who miss one appointment are significantly more likely to miss additional appointments.

Note

Any reference to a user in this documentation is to a user who is working on behalf of, or for a healthcare provider. A user isn't a patient.

Data is imported from a data source into Dynamics 365 Customer Insights, where predictions are generated by the AI model. These are surfaced in the unified patient view for care team members, and can also be used to support outreach campaigns.

Preparing the missed appointments feature includes the following steps:

• Connect your data to Customer Insights.
• Train the model on your data.
• Evaluate model performance.
• Connect the predictions to the unified patient view.

Intended use

• Intended use: This feature is intended to be used by healthcare providers to predict the likelihood that patients will miss their next appointment and take action to help them attend their appointment, which also includes guiding outreach to the patient and conveying the prediction to the patient's care teams.

• Non-intended use: The feature shouldn't be used outside the parameters highlighted in the previous section. This feature is meant to help improve the likelihood of appointment attendance and the predictive information shouldn't be used to penalize patients in any way, including overbooking appointments or terminating care. Misusing model predictions can result in unfair treatment of a patient. For example, overbooking appointments based on the prediction could lead to longer wait times for the patient or a specific community.

• Training the model: Care should be taken to train the model in ways that won't incur bias against patients or patient cohorts. You can consider the approaches described in Data prerequisites and Mitigate fairness risks.

• Not a medical device: This feature is

  1. not designed, intended or made available as a medical device.
  2. not designed or intended to be a substitute for professional medical advice, diagnosis, treatment, or judgment and shouldn't be used to replace or as a substitute for professional medical advice, diagnosis, treatment, or judgment.

• Interpretation of results: Because the model takes appointment details as input, the model results are applicable to a patient’s specific appointment, not a patient generally, and the results should be interpreted as such. Additionally, this feature is predicting the likelihood of a missed appointment based on the attributes listed in Model attributes. It isn't a guarantor of any individual’s actual behavior. The model’s predictions should be regarded only as providing information that may aid in predicting a missed appointment, not as prescribing an action to be taken. Having such information, a care team can improve their decision-making process by proactively reaching out to a patient to reduce the likelihood of an appointment being missed.

Limitations

Technical limitations

The model isn't pre-trained and will need to be trained by the user of a healthcare provider.

Preview limitations

The feature is currently in preview. Therefore, this feature is subject to change, and:

1. Is subject to separate Supplemental Terms of Use.
2. Isn't supported by Microsoft Support and isn't covered by service-level agreements (SLAs).
3. Is designed for evaluation purposes and shouldn't be rolled out to live production systems.

Prerequisites

Licensing and software prerequisites

Your organization needs the following licenses to be able to use this feature:

• Microsoft Cloud for Healthcare license.
• Dynamics 365 Customer Insights license.
• Dynamics 365 Customer Service license, if integrating with the unified patient view solution.
• Appropriate license(s), if integrating with another Microsoft Cloud for Healthcare first-party model-driven application (For more information, go to licensing guide).

Data prerequisites

See Model attributes for a list of required and recommended data attributes for your dataset. These attributes are based on the FHIR standard and the Microsoft Cloud for Healthcare data model.

Important

• For each of the required fields, ensure that at least 75% of the records have data before attempting to provision Customer Insights. If Customer Insights is provisioned before making sure the data has the required fields populated, you'll have to manually set up the AI model later in the process (by refreshing or recreating the model).
• The amount of data available to train and test your model is the most important factor for your model’s performance.

Using fewer data points creates a higher risk of overfitting, which means that while the performance of the model on training data may suggest good quality, the model might not generalize well to future inputs. This might negatively impact the performance of the model in production.

There's also a risk of underfitting, which means that the model isn't able to learn the patterns in the data and won't perform well.

We recommend using at least one year of historical data of approximately 10,000 patients as a starting point. This will help the model generalize well to future inputs and learn complex patterns in the data, which will reduce the chances of overfitting or underfitting. In addition, we recommend the following:

• Clean the input data before using it (avoid duplicate data and empty rows).
• Ensure that the data is representative of your existing and future patient pool.
• Ensure that the data contains:
  • A significant number of appointments missed (at least 10% of the entire set).
  • A range of values for each factor (for example, data for different types of appointments).
  • A significant amount of data for the demographic groups represented in your patient population (for example, patients with disabilities or chronic conditions).

Important

The missed appointments feature is likely to produce inconsistent or under-performing results for groups that aren't well represented in the training dataset. To mitigate the fairness risk of the model performing worse for specific groups, ensure that the data contains a significant amount of data for patients in these groups.

Mitigate fairness risks

AI and machine learning systems can display unfair behavior. One way to define unfair behavior is by its harm, or impact on people. There are many types of harm that AI systems can give rise to.

For guidance on how to build your training dataset to mitigate fairness risks, see the previous section on Data prerequisites.

For additional guidance about how to define and mitigate fairness risks, including how to ensure that your model is performing fairly after it’s been trained, go to Machine learning fairness, Microsoft’s AI fairness checklist, and also see Responsible AI resources.

Compliance

You may train the model on protected health information (PHI), and the view of the model predictions in the unified patient view may contain PHI. Access to the Dynamics 365 Customer Insights instance with the missed appointments prediction and any unified patient view forms that contain the predictions should be limited to those who have access to PHI. To manage access, multiple unified patient view forms can be created in such a way that only some users at a healthcare provider end can see the prediction. For more information, go to Set up unified patient view controls.

The missed appointments feature also uses automated means to evaluate data and make predictions based on that data, and therefore has the capability to be used as a method of profiling, as that term is defined by the General Data Protection Regulation (GDPR). Customers’ use of this feature to process data may be subject to GDPR or other laws or regulations. You're responsible for ensuring that your use of Dynamics 365 Customer Insights, including the missed appointments feature, complies with all applicable laws and regulations, including laws related to privacy, personal data, biometric data, data protection, and confidentiality of communications.

See also

What is Microsoft Cloud for Healthcare?
Get started with patient trends (preview)
Set up Customer Insights for patient trends (preview)
Create AI model for missed appointments (preview)
Set up unified patient view controls (preview)
Troubleshoot patient trends (preview)
Use patient trends (preview)

How do you document no show appointments?

Why is important to follow up with a patients?

What steps can and should be taken to avoid missed appointments?

What should not be documented in a medical record?