Change logic for KDIGO staging

[alistairewj]

• kdigo-creatinine.sql - now creates a row for every creatinine value, along with (1) the lowest previous creatinine in last 48 hours and (2) the lowest previous creatinine in the last 7 days
• kdigo-uo.sql - added the time of earliest UO measurement that was used to calculate the rate (so, if we calculated a 6 hour rate using a measurement 3 hours ago, this would be the time of the measurement 3 hours ago). This time is used in staging (see below).
• kdigo-stages.sql added - stages AKI at all time points (rather than once for 48 hours/7 days)
• kdigo-stages-48hr.sql - subselects from above query, and gets the highest AKI stage in the first 48 hours of a patient's stay
• kdigo-stages-7day.sql - subselects from above query, and gets the highest AKI stage in the first 7 days of a patient's stay

Some logic changes:

• baseline has been redefined from lowest value in the first 24 hours of the ICU stay to the lowest value in the last 7 days. This is more consistent with KDIGO, which states baseline "...is known or presumed to have occurred within the prior 7 days" - and mentioned by Chris in an issue.
• UO rate is only used if the rate was calculated over at least N/2 hours, where N=6, 12, 24 as appropriate

Regarding some open issues/PRs, this closes #486 (no longer relevant), closes #484, and refers to #476 (but not necessarily closed).

Contingency table:

aki_48hr	aki_creat_48hr	aki_uo_48hr	n	% of stays
0	0	0	10299	29.5
0	0		1647	4.7
0		0	25	0.1
0			25	0.1
1	0	1	2902	8.3
1	0	2	7543	21.6
1	0	3	2853	8.2
1	1	0	1484	4.3
1	1	1	788	2.3
1	1	2	2263	6.5
1	1	3	2037	5.8
1	1		460	1.3
1	2	0	250	0.7
1	2	1	90	0.3
1	2	2	401	1.1
1	2	3	503	1.4
1	2		95	0.3
1	3	0	134	0.4
1	3	1	34	0.1
1	3	2	150	0.4
1	3	3	702	2.0
1	3		185	0.5
1		2	14	0.0
1		3	11	0.0

Query for above table:

with t1 as
(
SELECT ic.icustay_id
, ic.gender
, ic.admission_age AS age
, w.weight
, h.height
, los_icu
, first_hosp_stay
, aki_48hr
, k4.aki_stage_creat as AKI_stage_48hr_creat
, k4.aki_stage_uo AS AKI_stage_48hr_uo
FROM icustay_detail ic
LEFT JOIN weightfirstday w USING(icustay_id)
LEFT JOIN heightfirstday h USING(icustay_id)
LEFT JOIN kdigo_stages_48hr k4 USING(icustay_id)
)
, t2 as
(
select aki_48hr, AKI_stage_48hr_creat, AKI_stage_48hr_uo
, count(*) as n
from t1
where age > 18
and first_hosp_stay
and los_icu > 1
group by aki_48hr, AKI_stage_48hr_creat, AKI_stage_48hr_uo
)
select aki_48hr, AKI_stage_48hr_creat, AKI_stage_48hr_uo, n
, round(cast(n AS NUMERIC)*100/SUM(n) OVER (), 1) as frac
from t2
order by 1, 2, 3;

I'd be interested in people's thoughts on this - in particular @JackieMium / @steph409 / @chris20881

[alistairewj]

changed the logic to use the actual duration of time we estimated UO over, rather than fixing it to 6/12/24 hours; expect this to increase the rate of AKI, but doesn't seem to change too much:

aki_48hr	aki_stage_48hr_creat	aki_stage_48hr_uo	n	frac
0	0	0	11930	34.2
0	0		1647	4.7
0		0	25	0.1
0			25	0.1
1	0	1	3124	9.0
1	0	2	7315	21.0
1	0	3	1228	3.5
1	1	0	1729	5.0
1	1	1	854	2.4
1	1	2	2553	7.3
1	1	3	1436	4.1
1	1		460	1.3
1	2	0	287	0.8
1	2	1	106	0.3
1	2	2	467	1.3
1	2	3	384	1.1
1	2		95	0.3
1	3	0	158	0.5
1	3	1	45	0.1
1	3	2	187	0.5
1	3	3	630	1.8
1	3		185	0.5
1		2	16	0.0
1		3	9	0.0