Anirban Basu, "Estimating Person-Centered Treatment (PeT) Effects 
using Instrumental Variables: An Application to Evaluating Prostate
Cancer Treatments", Journal of Applied Econometrics, Vol. 29, No. 4,
2014, pp. 671-691.

This study used the linked SEER-Medicare database
(http://healthservices.cancer.gov/seermedicare/).  The data are
confidential and can be purchased after signing a data-use agreement.
The dataset used for this study is called the 'pedsf' file from the
SEER-Medicare database for patients diagnosed with prostate cancer.
The specific inclusion/exclusion criteria used for this study are
described below:

Criteria                                           Excluded       Total

Patients on the PEDSF file (1995-2005 diagnosis)                  220,049

Site1 is not 619                                     15,028       205,021

First diagnosis is after 2002                         1,557       203,464

Not eligible to both Part A & Part B or eligible
for HMO for 1 year before diagnosis and the entire
follow-up period.                                   115,127        88,337

Age >=66 & <=89                                       5,057        83,280

Restrict to first diagnosis of cancer (multsite="1")     32        83,248

No. of primary side on first diagnosis =1
(numprims="01")                                       8,470        74,778

Restrict to clinically localized stage
(tstage in ("1", "2"))                               11,949        62,829

Medicare status code (med_stcd="10")                    183        62,646

ESRD (chr_esrd=="0")                                     33        62,613

Seq1 (seq1="00")                                      1,893        60,720

Choose surgery with radiation therapy                 1,413        59,307

Did not chose surgery or active surveillance         32,781        26,526

Unknown death flag (dod_flg in (0", "1"))                23        26,503

Must survive at least 6 months from diagnosis           350        26,153

Unknown grade                                            69        26,084

Unknown race                                            255        25,829

Unknown zip-code level characteristics                  744        25,085

Less than two PSA tests on active surveillance
in year following diagnosis                           7,201        17,884

Drop if diagnosis year =1995                          1,974        15,910

Lack of CMS activity                                      2        15,908

HRRs do not have enough patients to calculate
IV rates                                              2,413        13,495


Since patients' residence zipcode is considered to be a restricted
variable, it is not released with the public version of the dataset. We
manually requested the release of the zipcode information by writing
to each PI of the SEER registries and obtaining their permissions.