Address Confidence Scores

Leverage data confidence scores to provide accurate in-network addresses to end users

Ideon’s Address Confidence Scores provide a quantitative assessment of the accuracy of provider addresses. Developed through machine learning techniques and manually verified data, Ideon’s data model maintains scoring for 3 million unique individual providers. These scores are an important contextual datapoint that enables users to optimize integrations based on their unique user flows. Ideon's provider-network dataset leverages carrier provider directories as the source of truth; the confidence scores offer an additional analytical layer that evaluates the accuracy of the carrier directory address data.

The Address Confidence Scores assign a High, Medium, or Low score to each address, with over 19 million unique addresses scored. These categorizations reflect the likelihood that the provider is actually practicing at the given address. The below notes include an interpretation of the scores and corresponding test results against manually verified data.

• High: the provider is very likely to be practicing at the address. In testing, over 85% of addresses were confirmed as valid.

• Medium: a toss-up in terms of likelihood the provider sees patients at the address. In testing, roughly 50% of addresses were confirmed as valid.

• Low: the provider is very likely to not be practicing at the address. In testing, over 93% of addresses were confirmed as invalid.

Making Requests for Address Confidence Scores

To retrieve address confidence scores, include the address_confidence_score field in the premium_fields array of your search request to the "Provider Search" endpoint (API docs). When requested, the address_confidence_classification will be included for each address in the addresses array for every provider. This enum provides the classifications high, medium, low, or null.

Access to address confidence scores requires appropriate API credentials and permissions. Please ensure your API key has been provisioned with access.

Filtering by Address Confidence Scores in the Response

For most member-driven provider search applications, we recommend dynamic filtering logic that acts on both the addresses returned for a provider and across providers returned for a search. While in many scenarios simply filtering out Low scored addresses can be effective, this can limit search results in rural regions and overly bias networks from health plans with more accurate directories.  

Ideon’s recommended filtering can be implemented in two ways: on the provider level, based on the number and distribution of addresses returned for each provider, and on the search level, based on the matching providers and distributions.  

• Provider Level — for each provider returned in a search: 

  • If there are addresses scored as each of High, Medium, and Low OR each of Medium and Low, filter out the Lows. 

  • If all addresses are scored Low, leave them. 

  • Sort addresses displayed by Address Confidence Scores, with High-scoring addresses first. 

• Search Level – across all providers returned in a search: 

  • If there are enough providers with addresses scored as High to satisfy criteria for volume for selection, filter out providers with only Medium or Low addresses. 

  • If there are not enough providers with addresses scored as High, show remaining providers as needed. 

  • Sort providers displayed by Address Confidence Scores or include it as a factor in your sorting algorithm. 

An important thing to note is that not all provider addresses are scored (returned as null). Typically, this happens in scenarios in which data for that specific provider (as defined by NPI) is received from one and only one source. The recommended action is to treat these scenarios similar to Medium-scored addresses.

We also recommend considering a warning flag or notification for end users when displaying Medium and Low addresses. This can be an effective way to communicate that the provider may not be seeing patients at the address, despite the health plan directory reporting it as in-network.