Defining the sample population

Question

Do 30-day complication rates significantly differ between paraesophageal hernia repair (PEHR), Heller myotomy, and/or fundoplication patients who are discharged on post-op day (POD) #1 vs. POD #0?

\(H_0\): There is no difference in the composite outcome of 30-day morbidity between patients undergoing the aforementioned procedures discharged on POD #1 and POD #0.

Sampling population

The population to whom we would like our results to be generalizable:

Otherwise healthy patients undergoing routine elective paraesophageal hernia repair, Heller myotomy, or fundoplication.

Inclusion criteria

Discharge on POD #0 or POD #1
Age > 18
ASA 1-3
Independentally functional
Elective, non-emergent surgery
Not on a pre-operative ventilator
Wound classification 1 or 2
Not undergoing concurrent or additional procedures.

Primary procedures

43279 - Laparoscopy, surgical, esophagomyotomy (Heller type), with fundoplasty, when performed
43280 - Laparoscopy, surgical, esophagogastric fundoplasty (eg, Nissen, Toupet procedures)
43281 - Laparoscopy, surgical, repair of paraesophageal hernia, includes fundoplasty, when performed; without implantation of mesh
43282 - Laparoscopy, surgical, repair of paraesophageal hernia, includes fundoplasty, when performed; with implantation of mesh

Build SQL Query

Selecting the correct CPT codes

First, I will select for the correct CPT codes. We want only patients that are undergoing the above listed procedures. If a patient has additional CPT codes listed, they are excluded. However, they are included if the other CPT codes are one of the above four.

select caseid,
    logical_or(cpt = "43280") cpt_43280,
    logical_or(cpt = "43279") cpt_43279,
    logical_or(cpt = "43282") cpt_43282,
    logical_or(cpt = "43281") cpt_43281
from nsqip.puf_cpt
group by caseid
having not logical_or(cpt not in ("43280", "43279", "43282", "43281"))

Join to main table

Since this does not require any targeted procedural data, I will simply join it to the main table:

select * from (
select caseid,
    logical_or(cpt = "43280") cpt_43280,
    logical_or(cpt = "43279") cpt_43279,
    logical_or(cpt = "43282") cpt_43282,
    logical_or(cpt = "43281") cpt_43281
from nsqip.puf_cpt
group by caseid
having not logical_or(cpt not in ("43280", "43279", "43282", "43281"))
)
inner join nsqip.puf using(caseid)

Enacting inclusion criteria

In order to produce our sample population, I will enforce the inclusion criteria below:

df %>%
  dplyr::filter(tothlos < 2,
                asaclas < 4,
                fnstatus2 == "Independent",
                wndclas < 3,
                electsurg,
                !emergncy,
                !ventilat)

Creating predictor variables

We need to edit a few of our predictor variables to make them more usable.

df %>%
  dplyr::mutate(bmi = nsqipr::bmi(height, weight),
                insulin = tidyr::replace_na(insulin, FALSE),
                when_dyspnea = tidyr::replace_na(when_dyspnea, "None"),
                type_prsepis = tidyr::replace_na(type_prsepis, "None"))

Creating outcome variables

Because our primary outcome is a 30-day composite rate of morbidity, we need some good way to obtain a composite rate. The easiest is probably the Dindo-Clavien classification scheme. Everyone will then be categorized into “no morbidity”, “minor morbidity”, or “major morbidity”. “Death” will also be accounted for. A patient will be categorized into the highest category for which their outcomes qualify them.

df %>% 
  dplyr::mutate(dindo = nsqipr::dindo(.),
                dindo_class = forcats::fct_recode(factor(dindo), `None` = "0", `Minor` = "1", 
                                                  `Minor` = "2", `Major` = "3", `Major` = "4", 
                                                  `Death` = "5"))

Selecting complete cases

Because matching requires complete data sets, we must select only the variables we need and then select records with complete data.

df %>%
  dplyr::select(caseid:sex, inout:age, electsurg, anesthes, surgspec:smoke, dyspnea, 
                when_dyspnea, fnstatus2:ascites, hxchf, hypermed, renafail, dialysis,
                discancr:transfus, prsepis, type_prsepis, emergncy:asaclas, optime, 
                tothlos, supinfec, wndinfd, orgspcssi, dehis, oupneumo, reintub, pulembol,
                failwean, renainsf, oprenafl, urninfec, cnscva, cdarrest, cdmi, othbleed, 
                othdvt, othsysep, othseshock, othcdiff, returnor, dindo, dindo_class) %>%
  dplyr::filter(complete.cases(.))

This leaves us with 10127 patients in our analysis.

Creating the Sample