Creating the SQL Query

Study question

Do composite morbidity and mortality outcomes significantly differ between open distal pancreatectomy and minimally invasive distal pancreatectomy?

Hypotheses

\(H_0\): There is no significant difference in composite morbidity, mortality, and post-operative fistula rates.

\(H_a\): The rates of composite morbidity, mortality and fistula rates are significantly different between the two study groups.

Desired study population

The population to which our results are ideally generalizable:

Inpatients undergoing elective, non-emergent open or minimally-invasive distal pancreatectomy without reconstruction and irrespective of indication.

Required CPT codes

Distal pancreatectomy

48140 - Pancreatectomy, distal subtotal, with or without splenectomy; without pancreaticojejunostomy

Splenectomy

38128 - Unlisted laparoscopy procedure, spleen
38100 - Splenectomy; total (separate procedure)
38101 - Splenectomy; partial (separate procedure)
38102 - Splenectomy; total, en bloc for extensive disease, in conjunction with other procedure (List in addition to code for primary procedure)

Inclusion criteria (and relevant NSQIP variables)

Inpatient (inout)
Admitted from home (transt)
With available demographic data (age, height, weight)
Surgery performed by a general surgeon (surgspec)
Elective surgery (electsurg, emergnt)
Not with a preoperative ventilator, sepsis, or wound infection (ventilat, prsepis, wndinf)
ASA class < 5 (asaclas)
Wound class < 4 (wndclas)
Operative time > 20 minutes (optime)
Without reconstruction (pan_gastduo, pan_reconstruction)
Without metastatic disease (pan_mstage, discancr)
Performed via a minimally invasive or open approach (pan_approach)
Without open assist (pan_open_assist)

Possible explanatory variables (and relevant NSQIP variables)

Our possible explanatory variables are as follows:

Minimally-invasive or open approach (group)
Sex (sex)
Age (age)
BMI (bmi, height, weight)
Diabetes (diabetes)
Smoking history (smoke)
Pre-operative functional status (fnstatus2)
History of COPD (hxcopd)
Ascites (ascites)
History of CHF (hxchf)
Use of hypertension medication (hypermed)
Pre-operative dialysis (dialysis)
Pre-operative steroid use (steroid)
Pre-operative weight loss >10% of body weight (wtloss)
History of bleeding disorders (bleeddis)
ASA classification (asaclas)
Pre-operative chemotherapy (pan_chemo)
Pre-operatiove radiotherapy (pan_radio)
Intraoperative placement of drains (pan_drains)
Malignant versus benign histology (pan_malig_histologic, pan_benign_histologic)
T-, M-, and N-stage (pan_tstage, pan_mstage, pan_nstage)

Primary outcomes (and relevant NSQIP variables)

Composite morbidity outcomes (supinfec, wndinfd, dehis, renainsf, orgspcssi, oupneumo, urninfec, pulembol, othbleed, othdvt, othsysep, othcdiff, returnor, reoperation, reintub, failwean, oprenafl, cnscva, cdarrest, cdmi, othseshock, cnscoma, neurodef, dopertod)
Rates of post-operative fistula (pan_fistula, pan_fistula_type, pan_fistula_intervention, ddrainremoval)

Dindo-Clavien classification

Dindo-Clavien classifications provide a convenient way to create composite outcomes. These are created from the NSQIP variables via the following categorizations:

1: Superficial SSI (supinfec), deep SSI (wndinfd), wound dehiscence (dehis), renal insufficiency (renainsf)
2: Organ-space SSI (orgspcssi), post-operative pneumonia (oupneumo), urine infection (urninfec), pulmonary embolism (pulembol), post-operative bleeding (othbleed), deep venous thrombosis (othdvt), systemic sepsis (othsysep), C. difficile infection (othcdiff)
3: Return to the OR (returnor), any reoperation within 30 days (reoperation)
4: Reintubation (reintub), failure to wean off the ventilator for 48 hours post-operatively (failwean), renal failure requiring dialysis (oprenafl), CNS CVA (cnscva), cardiac arrest (cdarrest), myocardial infarction (cdmi), septic shock (othseshock), coma (cnscoma), new neurological deficit (neurodef)
5: Death (dopertod)

In addition, I will use targeted data regarding post-operative fistulas to help assign to a Dindo-Clavien classification:

2: If fistula intervention (pan_fistula_intervention) is listed as NPO-TPN, Spontaneous wound drainage, or Drain continued >7 days.
3: If fistula intervention (pan_fistula_intervention) is listed as Percutaneous drainage or Reoperation.

The patient is categorized into the highest category for which they qualify.

The Dindo-Clavien classifications can be further consolidated:

None - grade 0
Minor - grade 1 or 2
Major - grade 3 or 4
Death - grade 5

We will create a composite outcome called morbidity that is a Boolean variable:

\[f(x) = \begin{cases} 1, & \text{if major outcome}.\\ 0, & \text{otherwise}. \end{cases} \]

mortality will be an independently assessed Boolean variable:

\[g(x) = \begin{cases} 1, & \text{if death}.\\ 0, & \text{otherwise}. \end{cases} \]

Post-operative fistula

Post-operative fistula (pan_fistula) will be defined as any patient in the NSQIP database with one of the following:

pan_fistula_type listed as Grade B POPF or Grade C POPF.
pan_fistula_intervention listed as Percutaneous drainage, NPO-TPN, Reoperation, or Spontaneous wound drainage.
ddrainremoval (days until drain removal) recorded as greater than 21 days.

Secondary outcomes (and relevant NSQIP variables)

Post-operative amylase and post-operative day of highest value (amylase, pod)
Rates of delayed gastric emptying (pan_delgastric)
Rates of reoperation (reoperation)
Rates of readmission (readmission)
Length of hospital stay (tothlos)

Capturing relevant data from database

An SQL query to retrieve all of the above information from the NSQIP database can be structured as follows:

select * from

  (select * from

    (select * from

      (select * from

        (select puf.*, cpttab.pan_spleen from (
          select *,
          0 < countif(cpt in ("38129","38100","38101","38102") and primarysurg is true) over(partition by caseid) as pan_spleen
          from nsqip.puf_cpt
        ) as cpttab

        inner join nsqip.puf using(caseid)
        where cpt = "48140" and nproc = 1 and primarysurg is true
        )

      inner join nsqip.pan using(caseid, pufyear)
      )

    inner join (select as value array_agg(t order by amylase desc limit 1)[offset(0)]
      from nsqip.pan_amylase t
      group by caseid) using(caseid)
    )

  left join (
    select distinct(caseid), TRUE as readmission from nsqip.puf_readmission
    where readmission is true
    ) using(caseid)
  )

left join (
  select distinct(caseid), TRUE as reoperation from nsqip.puf_reoperation
  where reoperation is true
) using(caseid)

This query results in a total of 6239 patients.

outcome	OR	p
pan_fistula	1.39 [1.06, 1.83]	0.018
dindo_major	0.88 [0.65, 1.17]	0.371
dindo_death	1.40 [0.52, 3.76]	0.501
reoperation	0.82 [0.49, 1.40]	0.471
readmission	1.03 [0.82, 1.29]	0.781
pan_delgastric	0.72 [0.47, 1.11]	0.136

outcome	coefficient	p
log_amy	0.60 [0.40, 0.81]	<0.001
pod	-0.74 [-1.15, -0.33]	<0.001
tothlos	-1.40 [-1.71, -1.10]	<0.001

outcome	OR	p
pan_fistula	1.40 [1.06, 1.85]	0.018
dindo_major	0.88 [0.66, 1.18]	0.386
dindo_death	1.58 [0.50, 4.99]	0.432
reoperation	0.84 [0.49, 1.42]	0.512
readmission	1.04 [0.82, 1.30]	0.768
pan_delgastric	0.72 [0.47, 1.11]	0.134

outcome	coefficient	p
log_amy	0.59 [0.40, 0.79]	<0.001
pod	-0.73 [-1.14, -0.33]	<0.001
tothlos	-1.37 [-1.68, -1.07]	<0.001

Making the Match

Creating the SQL Query

Study question

Hypotheses

Desired study population

Required CPT codes

Distal pancreatectomy

Splenectomy

Inclusion criteria (and relevant NSQIP variables)

Possible explanatory variables (and relevant NSQIP variables)

Primary outcomes (and relevant NSQIP variables)

Dindo-Clavien classification

Post-operative fistula

Secondary outcomes (and relevant NSQIP variables)

Capturing relevant data from database

Cleaning the data

Removing patients with inadequate data

Applying inclusion criteria

Removing records with discrepancies in the data

Patients with both a malignant and a benign pathology

Patients with a benign histology and a T or N stage

Re-formatting explanatory variables

Creating primary outcome variables

Fistula

Dindo-Clavien outcomes

Creating secondary outcome variables

Selecting for only complete cases

Making the Match

Pre-weighting balance assessment

Outcomes prior to matching

Matching

Post-weighting balance assessment

Outcomes after matching

Modeling

Unadjusted models (`var ~ group`)

Adjusted models (`var ~ group + ...`)

Making the Match

Creating the SQL Query

Study question

Hypotheses

Desired study population

Required CPT codes

Distal pancreatectomy

Splenectomy

Inclusion criteria (and relevant NSQIP variables)

Possible explanatory variables (and relevant NSQIP variables)

Primary outcomes (and relevant NSQIP variables)

Dindo-Clavien classification

Post-operative fistula

Secondary outcomes (and relevant NSQIP variables)

Capturing relevant data from database

Cleaning the data

Removing patients with inadequate data

Applying inclusion criteria

Removing records with discrepancies in the data

Patients with both a malignant and a benign pathology

Patients with a benign histology and a T or N stage

Re-formatting explanatory variables

Creating primary outcome variables

Fistula

Dindo-Clavien outcomes

Creating secondary outcome variables

Selecting for only complete cases

Making the Match

Pre-weighting balance assessment

Outcomes prior to matching

Matching

Post-weighting balance assessment

Outcomes after matching

Modeling

Unadjusted models (var ~ group)

Adjusted models (var ~ group + ...)

Unadjusted models (`var ~ group`)

Adjusted models (`var ~ group + ...`)