National Cancer Institute Home at the National Institutes of Health | www.cancer.gov

SEER 13 Delay Model Adjusted for the Backlog of VA Cases in Submission Year 2011

A policy change of the Department of Veterans Affairs (VA) regarding data sharing on VA cancer cases resulted in underreporting on VA hospital cases for submission years 2007-2011. Correction factors to adjust for this underreportingExternal Web Site Policy (PDF) have been provided in the online version of the SEER Cancer Statistics Review.

Beginning with the 2009 submission of SEER data, some SEER registries began to report VA cases that had previously been withheld due to the VA policy changes. This caused a bolus of backlogged VA cases being reported in 2009, 2010 and 2011 that typically would have been reported in 2006-2008. This delay in reporting of VA cases represents a unique situation associated with the VA policy change and does not represent a reporting delay that would be expected to continue into the future. The SEER delay model is designed to adjust for cases that are reported to the registry after information for their diagnosis year is first made public and is modeled based on the past history of cases with delayed reporting. Since the VA policy change led to a one time short term increase in late cases, these excess cases in 2009, 2010 and 2011 should not be included in predicted the reporting delay factor for current and future years. This report described the methods used to readjust the VA cases received in 2009, 2010 and 2011 so that the bolus of cases received from the VA does not inappropriately influence the delay model. The adjustment to the VA cases described below are used to estimate the delay factors for current incidence estimates and will be used when fitting the delay model in future years.

As with the 2009 and 2010 submissions, to take into account of the VA backlog in 2011 submission in the delay adjustment model, the counts are adjusted by re-distributing VA cases in 2011 submission to previous submission years according to the expected counts from the delay distribution conditional on the current submission. Specifically, for each of the diagnosis year 2004 – 2008, given the total cancer count in submission year 2011, the proportion of cumulative cancer count in each subsequent submission year is calculated based on the estimated parameters from previous year’s reporting delay model. The VA cases in 2011 are re-distributed to each of the prior submission year according to this proportion. The adjusted total cancer count in that submission year was then calculated by combing the non-VA cases and the re-distributed VA counts.

Method Details

For diagnosis year i, Denote

  • graphic representing p1ij probability that a reported case is first reported in submission year j,
  • graphic representing p2ij probability that a previously reported case is removed in submission year j,

and

equation representing pij = p1ij - p2ij

Note that graphic representing pij represents the probability of a case in diagnosis year i is counted in submission year j.

For diagnosis years i = 2004 to 2008 and submission years j = i + 2 to 2011, let

  • graphic representing Aij number of reported cancers first reported in submission year j (adds),
  • graphic representing Dij number of previously reported cancers removed in submission year j (drops),
  • equation representing Nij = (Summation from k=i+2 to j of (Aik – Dik) total cancer count in submission year j,
  • graphic representing Vij number of cancers reported from VA hospitals in submission year j.

For diagnosis year i, given the total cancer count in submission year 2011, the proportion of cumulative cancer count in submission year j is

equation representing pij* = E(Nij)/E(Ni,2011) = ((Summation from k=i+2 to j of pik)/((Summation from k=i+2 to 2011 of pik)

Using the estimated parameters from previous year’s reporting delay model, we calculate graphic representing pij for diagnosis year i = 2004 to 2008 and submission years j = i + 2 to 2011, and consequently, we calculate graphic representing pij*.

For the registries that have VA count graphic representing Vij available, we verify that the net gain of VA counts is not greater than net gain of cases. If that happens, we replace the VA counts in the previous years such that

equation representing Vi,j-1 = Vi,j - (Ai,j - Di,j)
Then set equation representing Nij* = (Nij – Vij) + pij* × Vi,2011

graphic representing Nij* is then rounded to the nearest integer. graphic representing Nij* is the adjusted total cancer count in submission year j.

The adjusted “add count” graphic representing Aij* is calculated by

equation representing Aij* = Nij* – Nij-1* + Dij

Algorithm

The detailed adjustment algorithm is given below.

if graphic representing Vij is available then

do i = 2004 to 2008 /* Verify that the net gain of VA counts is not greater than net gain of cases*/ do j = 2011 to i + 3 If ( Vi,j - Vi,j-1 ) > ( Ai,j - Di,j ) then Vi,j-1 = Vi,j - ( Ai,j - Di,j ); end { do j = 2011 to i + 3 } /* calculate adjusted total count Nij*. */ do j = i + 2 to 2011 1) calculate pij*, using the estimated parameters from the reporting delay model. 2) Nij* = ( Nij – Vij ) + pij* × Vi,2011 (note that Ni,2011* = Ni,2011, since pi,2011* = 1). 3) round Nij* to the nearest integer. 4) if ( Nij* < Nij ) then set Nij* = Nij. end { do j = i + 2 to 2011 } /* make sure that adjusted add count Aij* is nonnegative. */ do j = 2011 to i + 3 by -1: if ( Ni,j-1* > Nij* + Dij ) then set Ni,j-1* = Nij* + Dij. end { do j = 2011 to i + 3 by -1 } /* calculate adjusted add count Aij*. */ When = i + 2, A_( i i+2 )^*=N_( i i+2 )^*. For j > i+2, do j = i + 3 to 2011: calculate the adjusted add count Aij*: Aij* = Nij* – Ni,j-1* + Dij. end { do j = i + 2 to 2011 } end { do i = 2004 to 2008 }