You are on page 1of 1

LETTER

NOTES & COMMENTS

The challenges of big data in dermatology made by experienced clinicians carefully reviewing
To the Editor: We read with interest the article by patient history, clinical examination findings, and
Kaffenberger et al highlighting the comorbid disease skin pathology and not by an objective value like
associations and impact of inpatient procedures on blood pressure or HgA1C. Because of this
patients hospitalized with pyoderma gangrenosum complexity, these diagnoses can be difficult to
(PG).1 We applaud the authors for recognizing the correctly identify in data sets.
prognosis of inpatients with PG is a subject that has If these guidelines are followed, the use of large
not been well described and deserves further administrative databases has the potential to play a
research. The results of this study were based on major role in advancing our understanding of rare
data from the Nationwide Inpatient Sample (NIS), a skin diseases. If nonvalidated approaches are taken,
20% stratified sample of all discharges from US these studies could undermine progress by present-
community hospitals, including individuals covered ing systematically biased and incorrect data. Big data
by Medicare, Medicaid, and private insurance and has the potential to make connections and find novel
those who are uninsured.2 All adult patients with an associations that are otherwise impossible to identify
ICD-9 [International Classification of Diseases, Ninth in smaller patient cohorts; the peril is that without
Revision] code for PG (686.01) as the primary or proper methods these insights might be erroneous.
secondary discharge diagnosis were included in the Megan H. Noe, MD, MPH,a and Arash Mostaghimi,
analysis. MD, MPA, MPHb
Although the research question is important, we
have concerns about the validity of ICD-9ebased From the Department of Dermatology, University of
identification of PG cases. Clinically, PG is a chal- Pennsylvania, Philadelphia, Pennsylvaniaa;
lenging diagnosis with a high misdiagnosis rate.3 The and Department of Dermatology, Brigham and
use of ICD-9 codes for the diagnosis of PG was Women’s Hospital, Harvard Medical School, Bos-
recently evaluated in the electronic medical records, ton, Massachusettsb
and only 45.3% of patients with $1 ICD-9 code for Funding sources: None.
PG were found to have PG after review of the
medical records.4 This suggests that more than half Conflicts of interest: None disclosed.
of the patients included in this analysis might not Correspondence to: Megan H. Noe, MD, MPH, 3400
actually have had PG. In the validation study, the Civic Center Blvd, Perelman Center for
positive predictive value increased significantly Advanced Medicine e 7 South, Philadelphia,
when patients had [1 code (69.2%) or when it was PA 19104
a dermatologist-rendered code (84.6%),4 suggesting
it’s possible to design an algorithm to identify PG E-mail: megan.noe@uphs.upenn.edu
within large, claims-based data sets. The NIS is
unfortunately not linked to the medical records,
and neither these advanced algorithms nor a
REFERENCES
database-specific validation can be performed. 1. Kaffenberger BH, Hinton A, Krishna S. The impact of underlying
While we encourage dermatologists to continue disease state on outcomes in patients with pyoderma gangre-
epidemiology and outcomes research for rare dis- nosum: a national survey. J Am Acad Dermatol. 2018. https:
eases within the field of dermatology, it is paramount //doi.org/10.1016/j.jaad.2018.02.007 [Epub ahead of print].
that our research aims are held to the highest 2. Overview of the National Inpatient Sample. Available at: www.
hcup-us.ahrq.gov/nisoverview.jsp#about. Accessed March 7,
standards. Ideally, validation of each database should 2018.
be done independently by using primary data. When 3. Weening RH, Davis MDP, Dahl PR, Su DWP. Skin ulcers
this is not possible, as with the NIS, researchers misdiagnosed as pyoderma gangrenosum. N Engl J Med.
conducting studies using administrative claims data 2002;347:1412-1418.
should employ validated algorithms to identify the 4. Lockwood SJ, Li DG, Butler D, Tsiaras W, Joyce C, Mostaghimi A.
The validity of the diagnostic code for pyoderma gangrenosum
primary outcome and related comorbidities. This is in an electronic database. Br J Dermatol. 2018. https:
especially critical when using large databases to //doi.org/10.1111/bjd.16446 [Epub ahead of print].
identify rare diseases that are based on clinical
evaluation. Correct dermatologic diagnoses are https://doi.org/10.1016/j.jaad.2018.03.059

J AM ACAD DERMATOL n 2018 e1

You might also like