Assignment of referring PCPs

[imccart]

We assign the referring physician in two ways: “Prior Visits:” This assignment looks at each operation and takes the most frequently visited physician (limited to E&M visits) by that patient in the prior year. Ties are broken by the most recent visit. But…we can identify frequency/recency of visits in two ways: 1) across all visits in the prior year; or 2) only among PCPs in the prior year. This choice naturally affects how many patients we are able to assign a referring physician and who that referring physician is. “Referring Physician Field:” This approach just looks to the physician listed as the referring physician in the claims data.

[imccart]

We observe 3,010,119 total operations over our estimation period from 2013 through 2018. Our referral coverage varies with different assignment processes:

We can assign a referring physician for 2,767,781 operations when we look at most frequent visits (regardless of whether they are a PCP). This increases slightly to 2,802,118 if we include the referring physician field.

We can assign a referring physician for 1,353,764 when we remove non-primary care specialties after calculating frequency/recency. This increases to 1,585,252 when we include the referring physician field.

If we focus only on PCPs from the beginning, we can identify a referring physician in 2,405,761 based on most frequent/recent visits, increasing to 2,444,494 when including the referring physician field.

Note that the referring physician listed in the claims is much more likely to be a surgeon. Specifically, that field is populated for 2,000,798 operations in our data, and the referring physician is a surgeon in 1,355,420 of those. The referring physician is specifically an orthopedic surgeon in 1,324,724 of those. So the referring physician explicitly listed in the claim is another orthopedic surgeon in around 65% of cases. In cases where the referring physician is a surgeon, the referring physician and operating physician are almost always the same (95%). Looking at the same practice this percentage increases to 97.5%.

So unsurprisingly, we get the highest coverage when we either impose no restrictions on the referring physician specialty or when we impose a restriction before measuring frequency/recency of visits. But imposing no restrictions is not ideal because it will tend to introduce a lot of noise (capture “referrals” that weren’t really referrals).

[imccart]

If we impose PCP restrictions before measuring frequency/recency, we tend to decrease the number of visits associated with the referring physician. The mean number of visits in this case is 4 with a median of 3. The 25th percentile of visit frequency is 2 and the 10th percentile is 1. So nearly 25% of observations are being assigned a referring physician based on just 1 visit. The distribution shifts up slightly when we look at overall frequency/recency or when we impose PCP restrictions after calculating frequency. In both of those cases, the mean number of visits is around 5, the median is around 4, and singular visits account for less than 10% of referrals.

Although this may not matter given the uninformativeness of the referring physician field, we can look at the match of physician NPIs between our two assignment methods. Here, we obtain the highest match rate when we impose the PCP restriction after measuring frequency of visits. This yields a match rate of 27% compared to 23% (focusing only on PCPs from the beginning) and 15% (without any PCP restrictions).

[imccart]

Given the preponderance of same-physician and same-practice referrals, we should prioritize assignment of referring physician based on frequency/recency. Argument is that referring physician field is uninformative.

That still leaves a few decisions: 1) measure frequency/recency based only on PCPs: 2) measure frequency/recency overall and then limit to PCPs; and 3) whether to impose restrictions on count of visits. Identifying based on all physicians and then limiting to PCPs (so we only identify the referring physician if they are a PCP and they are the most frequent/recent visited physician for that patient) will miss a lot of operations. Another approach would be to look only at PCPs from the outset, but with some restriction on the count of visits.

[imccart]

Current approach:

• Ignore referring physician field
• Restrict to PCPs from the outset and limit to >=2 visits. In the code right now, this is indicated by PCP_First=1, PCP_Only=1, and PCP_Referring=0.
• Reasonable backup is to restrict to PCPs after identifying frequency/recency physician. In the code, this is indicated by PCP_First=0, PCP_Only=1, and PCP_Referring=0. Need to update this combination of settings as of March 2023.