Avoiding all analyses of the 'effectiveness' of ICU (i.e. effect of decision to admit, or early admission) Factors affecting outcome for patients not admitted Examining the effect of occupancy - in the frailty model allows for confounding by hospital - but removes the effect of hospital that operates through occupancy Recast results to emphasise interest in outcomes of patients referred to and not immediately admitted (esp. those with a recommendation for critical care)

inbox: - consistency - ward assessment not bedside - occupancy - sensitivity - by occupancy definition - by data quality - need to update doit - be consistent and use ICNARC physiology points through-out and NEWS for classes - formatting for the BMJ - sensitivity analyses - check thresholds - add a variable representing delayed discharges to the model @later @idea - run survival model at 7d as single level - bootstrap CI - check/read about which SE to use after bootstrap (simple vs ...) at present assuming central limit theorem holds and sampling by bootstrap creates this distn then use Z

now: - [ ] TODO(2016-06-02): table edits as per DH - [ ] TODO(2016-06-02): methods edit re recommend as per KR - [ ] TODO(2016-06-02): submission to BMJ

paper: abstract: intro: methods:

results:
	- table 1: update
		- mortality at 1 year?
	- determinants of decision to admit
		- convert predicted additional	admissions to percentages
			 We estimated that in this sample had there been no limitations on capacity then an additional 122 patients (95\%CI 53-186) would have been admitted.
	
	
	
	- add in mortality in first 7 days vs 90 days - perhaps to supp figure caption





discussion:
	you are going to write about how the factors in play at the bedside assessment affect decision making, care and outcome
	you can safely leave the causal pathway of	decision making acting through delivery of care to the IV paper 

figures:
	
	
	

tables:
	
	
	- add early admission and decision to admit into the survival model? @today
	- table 4 
		add as additional column to table 3?
	
	
	- supplementary
	
	- maybe
		- table 1 repeat but comparison of accept vs refused vs limits
		- table 2 repeat (decision effects)
			- effect of recommendation
			- effect of decision

BMJ submission:

• Title - All manuscripts
• Names, addresses, and positions of all authors - All manuscripts
• Copyright - All manuscripts
• A statement that "the Corresponding Author has the right to grant on behalf of all authors and does grant on behalf of all authors, an exclusive licence (or non exclusive for government employees) on a worldwide basis to the BMJ Publishing Group Ltd to permit this article (if accepted) to be published in BMJ editions and any other BMJPGL products and sublicences such use and exploit all subsidiary rights, as set out in our licence."
• Competing interest statement - All manuscripts
• All authors have completed the Unified Competing Interest form atwww.icmje.org/coi_disclosure.pdf (available on request from the corresponding author) and declare: no support from any organisation for the submitted work [or describe if any]; no financial relationships with any organisations that might have an interest in the submitted work in the previous three years [or describe if any], no other relationships or activities that could appear to have influenced the submitted work [or describe if any].- Please note: The corresponding author must collect Unified Competing Interest forms from all authors and summarise their declarations as above within the manuscript. You do NOT need to send copies of the forms to the BMJ. Please click here for further guidance
• Details of contributors, and the name of the guarantor - All full papers (eg not fillers or personal views)
• Transparency declaration: a statement that the lead author (the manuscript’s guarantor) affirms that the manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned (and, if relevant, registered) have been explained
• Details of ethical approval (or a statement that it was not required) - All research studies
• Details of funding - All research studies
• Details of the role of the study sponsors - All research studies
• Statement of independence of researchers from funders - All research studies
• Trial registration details (registry and number) - All clinical trials
• Data sharing statement All research studies
• As an appendix (or in a supplementary file)
• CONSORT statement - All randomised controlled trials PRISMA statement - All systematic reviews EVEREST statement - All economic evaluations STARD statement - All diagnostic research papers STROBE statement - All observational studies
• Patient consent - All articles that include photographs or descriptions of patients

Checklist -Article requirements

http://www.bmj.com/about-bmj/resources-authors

• Title - all manuscripts - Title page

  • Names, addresses, and positions of all authors plus email address for corresponding author, ensuring that all people listed as authors fulfil the criteria for authorship - all manuscripts

• Copyright/licence for publication - all manuscripts

• A competing interest declaration - all manuscripts

• Details of contributors and the name of the guarantor - all original research and education articles Signed patient consent forms - all manuscripts with personal information about a patient

• Statements regarding ethics approval; informed consent from participants; funding; the role of the study sponsor in study design and the collection, analysis, and interpretation of data and the writing of the article and the decision to submit it for publication; the independence of researchers from funders and sponsors; and the access of researchers to all the data - all original research articles

• An observational study please follow the STROBE guidelines and submit as a supplemental file the study protocol, if there is one

  • The registration details, if the study has been registered - these should be added to the last line of the paper's abstract. We will also ask for claification about whether the study was registered before data acquisition or analysis began.
  • The protocol, if one exists - uploaded as a supplemental file to the submitted paper.
  • A clear statement of whether the study hypothesis arose before or after inspection of the data (and, if afterwards, we will need an explanation of steps taken to minimise bias).
  • A completed STROBE checklist - uploaded as a supplemental file to the submitted paper. We will pay particular attention to these items which ask authors to "explain the scientific background and rationale for the investigation being reported" and "state specific objectives, including any prespecified hypotheses."

@later: - writing - model decision to admit as binary excluding patients with a treatment limitation order - tidying tasks - re-run with SOFA score to keep severity adjustment consistent - exclude patients with level 3 support and replace v_ccmds_now with organ_support (soup) - redo exact OR in text after updating model - redo MOR excluding patient level factors - sensitivity models - full cohort without rxlimit exclusion - sensitivity using time2icu as the dependent var - early admission as the dependent var instead - within subgroup with a recommendation for critical care - comment in text on - effect of occupancy - number of additional patients who would be predicted to be admitted were occupancy OK (overall, and amongst those recommended) get at the weighted mean by taking n.extra and dividing by the study pop - full model - in subgroup recommended (this might be the model to present) - inter site variability via MOR - translate into predicted admissions too?

	- finish with mortality model for decision to admit 
		- for all patients 
		- for patients recommended to critical care
		show similar outcomes, in discussion then comment on incomplete adjustment which sets up next paper
	- table comparing the three pathways
	- incidence table
				- incidence model and table
						- re-run as 12 hourly model to allow calculation of shift	incidence (out of hours vs in hours) @later

FIXME: 2015-11-27 - [ ] problem with stata connecting via ODBC; seems to hang; for now all prep files drop this and just copied in completed work FIXME: 2015-11-27 - [ ] problem with waf; seem to need to run it twice for it to work

2015-11-27

• now rewriting as per BMJ submission
• maybe introduce the ideas of patient groups to follow as per the rewrite in progress for paper-spotearly

140509

• there are patients in the sample that were admitted to ICU after one week (because they were permitted during the match) but I think my original intention was to match them but then replace their ICU admission time as missing

140506 TRY - drop survival errors early so that you have the same number of patients in the initial analyses as you do in the survival analysis

140506

• Thought about include occupancy of ICU in severity models (at least to show that this doesn't have an effect)? No -- leave this out for now: this should go into the survival and delay paper
• converted output of ts_model_count_combine_news.do from xls to csv

140505

• ts_model_ward_severity.do previously called tb_model_ward_severity.do: renamed
• modified ts_model_ward_severity.do so that it writes to a csv file which is then linked to the tables spreadsheet

140413

• added outsheet commands to table files in order that I can generate tables for publication from data
  • table 1a completed
  • table 1b completed
  • supp table 1 - site chars - completed
  • supp table 2 - incidence by news risk class - completed
• replaced all spot_ward references with mas_spotepi (incl those in vcode/spot_ward) so that I don't forget when working in the future with new transfers
• created wr08_sensitivity analysis

140328

• created a single record per patient survival data set by modifying cr_survival

• go back to estimating the median odds ratio; according to this answer it is not possible to derive a confidence interval for this (easily) in stata

• median hazard ratio??

  • So to estimate the MHR you now need to local twoinvtheta2 = 2 / (e(theta)^2) local mhr = exp(sqrt(2*e(theta))*invF(twoinvtheta2',twoinvtheta2',0.75)) di "MHR: `mhr'"

• produced relative survival estimates using strel2 from LSHTM

• combined all this code into wr06_mortality.do

140324

• added in mortality section wr06_mortality.do
• ts_model_count_news_combine completed but stops before producing latex table and the output goes directly into a spreadsheet for formatting

140318

• re-run tb_model_ward_severity.do: works much better and figure no longer has Ulster as an outlier
• recreated hes_providers table in spot db after correcting the total admissions for Northern Ireland sites in the original spreadsheet
  • this meant re-running import_excel.py, index_table.py, and make_table.py
    • then re-creating the derived tables sites_within_hes and sites_early
  • then running cr_sites.do

140313

• corrected the tb_model_count_news_high model so that days without admissions are included
• corrected sites_early table by re-running make_table.py for

• new system for analysis files: use wrXX_ prefix for results paragraphs in paper
• similarly

• added back in all cr_* files. These will allow creation of the data set.
• I have not actually re-created all the data via SQL and Python code in /ccode
• source table in SQL is spot_early.working_early - connecting and downloading this table creates working_raw.dta
• STROBE diagram data checked and all working up to cr_working.do

These are notes on the analysis to be performed for the spot-epi paper. The code here will largely come from the code used for spot_ward. In fact, it the directory is a branch published from the spot_ward repo. For now, I have moved all files into a subfolder called spot_ward so that I have a fresh start. I will then move the relevant files out one-by-one.

Archive: - drop patients with rxlimits from second half @done(2016-03-02) @project(abstract) - 95%CI for delay when refusing admission @done(2016-03-02) @project(abstract) - 95%CI for delay with high occupancy @done(2016-03-02) @project(abstract) - refocus using themes in Chen VA paper and Stelfox paper @done(2016-03-02) @project(intro) - explain concept of tracking patients not admitted @done(2016-03-02) @project(intro) emphasise USP that we have followed all patients - explain why mortality analysis in subgroup of patients rejected @done(2016-03-02) @project(methods) - add in @done(2016-03-02) @project(methods) Cox proportional hazards were used to model survival with a shared frailty (random effect) for hospitals which was is reported as a Median Hazard Ratio (MHR).[Bengtsson and Dribe, 2010, #60939] The proportional hazards assumption was checked by inspecting plots of smoothed exponentiated standardised Schoënfeld residuals, and re-entering terms using time-varying co-efficients where necessary. - mortality by treatment decision? @done(2016-03-02) @project(results) - add back in incidence per 1000 admissions @done(2016-03-02) @project(results) see old working - sort out table and figure numbering @done(2016-03-02) @project(results) - comments to move to the discussion @done(2016-03-02) @project(results) - suggesting that, unless there are unmeasured patient level risk factors more important than those already measured, this variability is not due to incomplete risk adjustment. - add decision making into survival model @done(2016-03-02) @project(results) not because you are interested in its effect, but because you wish to examine the MHR after adjusting for decision - and whether or not that is how 'site' contributes to outcome - estimate model - report/comment if makes a difference to MHR - add note into results re this - figure 1 (strobe plus pathways) @done(2016-03-02) @project(figures) - a strobe - b recommended - c decision - figure 2 (time2icu) @done(2016-03-02) @project(figures) borrow from spot early code - supplementary @done(2016-03-02) @project(figures) - incidence and case finding - daily hazard - severity of illness and outcome - critical care occupancy - schoenfeld_residuals_icnarc_score.jpg - maybe @done(2016-03-02) @project(figures) - time 2 icu figure - table 1 (baseline characteristics) @done(2016-03-02) @project(tables) - modifications - with age categories - peri-arrest status - assessment timing - table 2 (occupancy effects) @done(2016-03-02) @project(tables) - convert to tables 3a and 3b @done(2016-03-02) @project(tables) - table 3 (determinants of decision to admit) - table 4 (determinants of prompt admission) - incidence table @done(2016-03-02) @project(tables) - caption The bottom line shows the monthly incidence of patients categorised by NEWS Risk Class referred to, and assessed on the ward by critical care. Above this, incidence rate ratios (IRR) with 95%CI are reported for hospital, and timing factors. - bed pressure: high/medium/low @done(2016-03-02) @project(inbox) - fluctuates with season of the year: need figure to back this up @done(2016-03-02) @project(inbox) - add follow-up metric @done(2016-03-01) @project(inbox) - convert -- to 'to' @done(2016-03-02) @project(inbox) e.g. (rate ratio 1.82, 1.35 to 2.46) (≥12 hours: rate ratio 1.77, 1.07 to 2.93) - convert incidence model to per 1000 admission metric @done(2016-03-02) @project(inbox) - double check occupancy order in markdown now you have re-levelled room_cmp2 @done(2016-03-02) @project(inbox) - check levels of age category predictors in models (k1 not k0 as baseline) @done(2016-03-02) @project(inbox) - redo predictors @done(2016-02-09) @project(@next) - add in time-varying covariates to survival models @done(2016-02-09) @project(@next) - drop site level predictors except from incidence model @done(2016-02-09) @project(@next) - simplify site level predictors (drop ?emergency admissions, CMP throughput) ... or just remove through out - switch to ccmds 2 or 3 (now) (back from osupp) @done(2016-02-09) @project(@next) @continue: @done(2016-02-09) - one tricky later paragraph @done(2016-02-09) - see @resume tag below @done(2016-02-09) - tidy up decision models to use cleaner set of predictors as per mortality model - add the code @done(2016-01-29) @project(results) - supplementary figure for schoenfeld residuals @done(2016-01-28) @project(results) - run survival model in full cohort @done(2016-01-27) @project(results) - report multi-level and single level model @done(2016-01-27) if the effect of occupancy is due to site then the multi-level model may hide this is occupancy is strongly collinear with site need to try and fit an interaction between site and occupancy probably too complicated so just report both models - and pick one for the main results and one for the ESM - recommended subgroup @done(2016-01-27) @project(results) - all @done(2016-01-26) @project(results) - check @done(2016-01-26) @project(results) - check @done(2016-01-26) @project(results) - check @done(2016-01-26) @project(results) - redo with centred age cat @done(2016-01-27) @project(results) - all accept - recommend accept - check @done(2016-01-26) @project(results) - redo with centred age cat @done(2016-01-27) @project(results) - check @done(2016-01-26) @project(results) - redo with centred age cat @done(2016-01-27) @project(results) - check @done(2016-01-27) @project(results) - move age cat to 18-39 then don't need to redo the others @done(2016-01-27) @project(results) - run survival model with icu_accept? @done(2016-01-27) @project(results) probably not since this implies assessment of the decision - add in time-varying component to acute severity score @done(2016-01-26) @project(results) - run a survival model without patient level covariates to report stability of MHR @done(2016-01-27) @project(results) - consider looking for an occupancy severity interaction @done(2016-01-27) @project(results) - occupancy univariate effects table @done(2016-01-22) @project(results) - ensure matches structure of other models @done(2016-01-22) @project(results) - run MHR bootstrap (since don't need competing risks) @done(2016-01-22) @project(results) - fix errors in recommended subgroup table @done(2016-01-22) @project(results) - tables @done(2016-01-18) @project(results) - table 1 @done(2016-01-18) - sfig 1: occupancy over time @done(2016-01-18) @project(results) - tidy up vars to match limited subset @done(2016-01-18) @project(results) - command line option for sims and subgroup @resume @done(2016-01-19) @project(results) - for all patients @done(2016-01-19) @project(results) - for those recommended @done(2016-01-19) @project(results) - siteonly @done(2016-01-19) @project(results) - effect of occupancy @done(2016-01-19) @project(results) - inter-site variability @done(2016-01-19) @project(results) - comparison by using model without patient level vars @done(2016-01-19) @project(results) - all @done(2016-01-19) @project(results) - delay to admission models @done(2016-01-22) @project(results) for the purpose of demonstrating that decision making is key hence need to include icu_accept - convert to updated set of predictors @done(2016-01-21) - add back in decision and work with - full cohort - recommended subgroup - - effect of occupancy - intersite variability cannot get at this in a competing risks model - could use early4 in a logistic regression to estimate this, but capture the hazard ratios from the main competing risks models - build early4 model as per accept - adjust code @done(2016-01-21) - update doit @resume @done(2016-01-22) ✔ start work converting participating hospitals to Rmarkdown @resume(2016-01-01) @done (16-01-01 18:18) @project(@continue) - occupancy paragraph @done(2016-01-13) @project(@continue) - occupancy table @done(2016-01-13) - occupancy figure @done(2016-01-12) - incidence models paragraph @done(2016-01-13) @project(@continue) - incidence table - ta_model_count_news_all @done(2016-01-12) - tb_model_count_news_high @done(2016-01-12) - incidence figure @done(2016-01-12) ✔ patient characteristics paragraph @done (15-12-31 11:47) @project(inbox) ✔ table 1 @done (15-12-31 12:28) @project(inbox) ✔ supp figure 1 (mortality pattern) @done (15-12-31 12:49) @project(inbox) ✔ supp figure 1 (severity - dead7 association) @done (15-12-31 13:00) @project(inbox) + done this today (70 patients) and corrected the text and the ICU admission numbers @project(FIXME) + see working_survival_single.dta @project(FIXME) + tried then running the laplace command in stata; nice because it directly estimates percentile survival; but I can't seem to pull out a random effect estimate for the site effect (fixed only) @project(FIXME) + exp(sqrt(2Vinvnormal(0.75)) where V is the random effects variance when the random effects are normally distributed @project(FIXME) + need to transform the above equation to work with the gamma distribution since this is used for the random effects in stata @project(FIXME) + frailty in stata has mean 1 (assumed) and variance \theta (which is estimated from the data) @project(FIXME) + the invnormal(0.75) term now needs to be replaced with the upper quartile from the an F distribution which has a numerator and denominator (2V,2V) (see ref [Bengtsson and Dribe, 2010, #60939] appendix) @project(FIXME) + sites_within_hes @project(FIXME) + sites_within_cmpd @project(FIXME) + sites_early @project(FIXME) Now includes HES data for Wales and Northern Ireland + ts_ prefix for "table supplementary" @project(FIXME) + tb_ prefix for "table" @project(FIXME) + fs_ prefix for "figure supplementary" @project(FIXME) + fg_ prefix for "figure" @project(FIXME) - skim and check methods @done(2015-12-19) @project(@continue) - incidence figures from GEE model @done(2015-12-29) @project(results) - report model as supplementary table? - remove elective emergency indicator - swap beds_none for room_cmp2 - remove referral pattern adjustment - add back in weekend - remove mp_throughput - check incidence per 1000 is for NEWS high risk ? @done(2015-12-29) - basic @done(2015-12-29) @project(results) - add in major groups @done(2015-12-29) @project(results) - write abstract @done(2015-12-18) - report occupancy and decision from time2icu model - report mortality models - run mortality model in cohort not recommended without limits this asks the qn: who should be offered critical care esp if you use combined endpoint of death or admission to critical care in the next week - model code @done(2015-12-18) @project(Mortality models) - overall 7 and 90 day survival models @done(2015-12-18) @project(Mortality models) - subgroup @done(2015-12-18) @project(Mortality models) - model outputs @done(2015-12-18) @project(Mortality models) - sanity check: add in report of model w/o frailty @done(2015-12-18) @project(Mortality models) - why missing 'p' (b/c only 1 sim) @done(2015-12-18) @project(Mortality models) - recommended and refused @done(2015-12-18) @project(Mortality models) - refused @done(2015-12-18) @project(Mortality models) - interaction of recommendation? @done(2015-12-18) @project(Mortality models) probably can drop this since you are looking at within subgroup effects - frailty / site level variation @done(2015-12-17) @project(Mortality models) - use re-sampling code - but is v slow; will need to run overnight? @done(2015-12-18) @project(Mortality models) - model time to admission for those recommended with MHR etc as before @done(2015-12-16) - next steps - - would need to include competing risk of death since censoring is not independent @done(2015-12-16) - fit simplified version of model @done(2015-12-15) - add in frailty @done(2015-12-15) - examine impact of 'correct decision' (i.e. what is the cost of being refused if admitted in the end) @done(2015-12-15) - just do simple model with CR but not frailty since combination is difficult @done(2015-12-16) - extract MOR @done(2015-12-10) - use data set up structure from paper-spotearly @now @done(2015-12-10) + copied the prep folder and made the same @done(2015-12-10) - include occupancy @done(2015-12-08) - convert from room_cmp to room_cmp2 @done(2015-12-08) - exclude patients with Rx limits @done(2015-12-08) - basic summary in text @done(2015-12-02) ✔ descriptive figure @done (15-12-02 13:38) - include description of patient groups as per draft from paper spot early @done(2015-12-04) - show differences via stream graph which allows inspection of mortality @done(2015-12-04) - facet stream graph by occupancy - repeat for those recommended for critical care - switch to waf or make file structure @done(2015-11-27) + copied data folder and made the same @done(2015-11-27) + symlinked the project_paths files from bld - fix tb_model_count_news_high @done(2015-12-01) - problem with number of sites @done(2015-12-01) - check constant for IRR @done(2015-12-01) - place under waf control @done(2015-12-01) * - [ ] TODO(2015-12-01): test code for figure count_news_high @done(2015-12-01) - place under waf control @done(2015-12-01)

I