﻿========================================================================================

CONCURRENT FRACTION UNDER DIFFERENT EXPOSURE-WINDOW DEFINITIONS

========================================================================================

Primary cohort: 2365 patients, 727 meeting the renal criterion at any point.



A patient is 'concurrent' when the first qualifying creatinine falls at or

before the end of the exposure window, so nothing about that patient was

predicted. The renal criterion is held fixed at our modified KDIGO

definition throughout; only the exposure window moves.



 exposure_window_end_h  patients_meeting_criterion  already_met_by_window_end  concurrent_pct  first_met_after_window_end

                     6                         727                         49             6.7                         678

                    12                         727                        248            34.1                         479

                    18                         727                        454            62.4                         273

                    24                         727                        531            73.0                         196

                    36                         727                        596            82.0                         131

                    48                         727                        665            91.5                          62

                    72                         727                        693            95.3                          34



========================================================================================

SAME GRID, BY PEAK LACTATE STRATUM

========================================================================================

Reported because the concurrent fraction is itself strongly related to the

exposure, which is the mechanism by which window overlap inflates an

apparent association.



lactate_stratum  patients  meeting_criterion  concurrent_pct_at_6h  concurrent_pct_at_12h  concurrent_pct_at_24h  concurrent_pct_at_48h

          <=1.9       815                163                   5.5                   27.6                   62.0                   86.5

        2.1-3.0       570                143                   4.9                   26.6                   65.7                   87.4

        3.8-5.9       765                293                   7.8                   37.5                   76.5                   93.2

     8.6 capped       215                128                   7.8                   43.0                   87.5                   98.4



========================================================================================

HOW THE ASSOCIATION CHANGES WHEN CONCURRENT EVENTS ARE LEFT IN

========================================================================================

The contrast between an analysis that counts every event and one that

counts only events arising after the exposure was complete.



 exposure_window_end_h  no_landmark_OR no_landmark_ci  no_landmark_events  no_landmark_n  landmark_OR landmark_ci  landmark_events  landmark_n  ratio_of_ORs

                    12            6.31      4.46-8.93                 727           2365         5.19   3.49-7.72              479        2117          1.22

                    24            6.31      4.46-8.93                 727           2365         2.18   1.16-4.07              196        1834          2.90

                    48            6.31      4.46-8.93                 727           2365         0.59   0.13-2.64               62        1700         10.68



The ratio of odds ratios quantifies how much of the apparent association

an overlapping-window design would attribute to prediction in this cohort.

It is a property of these data, not an estimate of bias in any published

study, and should be presented as such.



Outputs: <local path>\analysis\results\manuscript_v3

