Calculate outcomes

Calculate outcomes#

This notebook calculates outcomes based on travel times from Lower Super Output Areas (LSOAs) to nearest (in time) acute stroke units that provide thrombolysis and/or thrombectomy.

This notebook calculates the outcomes using the stroke outcome model. The following notebooks combine different cohorts of patients to find the results for a mixed population, and then further notebooks plot the results.

Plain English summary#

We would like to know how travel time to stroke units affects patient outcomes for different parts of England and Wales. We can explore this by keeping the same timings for everything else between the start of the stroke and the beginning of treatment and only adjusting the travel time. For example, a patient who lives 20 minutes further away from the stroke unit than another patient will have an onset to treatment time 20 minutes longer.

In this notebook we look at each Lower Super Output Area (LSOA), a small region of the country that contains a couple of thousand people. For an idea of scale, the city of Exeter is split into around 70 or 80 LSOA. In a more rural location, a larger geographical area must be covered to collect a couple of thousand people and so the LSOAs are larger.

We find a few travel times:

the travel time from each LSOA to its nearest stroke unit
the travel time from each LSOA to its nearest thrombectomy unit
the travel time between each LSOA’s nearest stroke unit and its nearest thrombectomy unit, for transfers between hospitals

We then add these to fixed stroke pathway times to calculate a few times to treatment:

time to IVT when travelling directly to the nearest stroke unit
time to MT after a transfer from the nearest stroke unit to the MT unit
time to IVT when travelling directly to the MT unit
time to MT when travelling directly to the MT unit

In the UK most of the stroke units will provide thrombolysis but not thrombectomy. Some patients will live nearest to an MT unit and so both of their sets of times will be identical. For most patients though, the time to IVT will be longer if travelling to the MT unit because the unit is further away than their nearest unit. The time to MT is usually longer if a transfer is required than if the patient went directly to the MT unit to start with, partly because the patient spends more time travelling between units and partly because there are delays in getting the patient ready to travel between hospitals.

We can then feed these treatment times into the stroke outcome model. This will give us two sets of outcomes for patients in each LSOA in England and Wales: one set for patients who travel directly to the nearest stroke unit, and a second set for patients who travel directly to the MT centre.

Aims#

Calculate outcomes after treatment for these cohorts:

nLVO with IVT
LVO with IVT only
LVO with MT only
LVO with both IVT and MT

Calculate outcomes after treatment for these timings:

IVT at the nearest stroke unit
MT after transfer to the MT unit
IVT after redirection to the MT unit
MT after redirection to the MT unit

Then save the outcome data to file for later use.

Two sets of outcomes will be saved:

population-averaged added utility, shift in mean mRS, proportion with mRS<=2
mRS distribution after treatment

Method#

The travel times are calculated from each LSOA centroid to the location of each stroke unit.

Travel times are estimated using Open Street Map[1], accessed by routino [2].

[1] https://www.openstreetmap.org/

[2] https://packages.debian.org/sid/routino

load in stroke-map package
Calculate travel time from each LSOA to its nearest acute stroke unit and to its nearest unit providing thrombectomy
Add on fixed pathway timings (onset to call, ambulance on scene, door-in-door-out etc.) to find time to thrombolysis and time to thrombectomy
Feed treatment times into the outcome model to find outcomes for each patient subgroup at those treatment times.

Load packages#

import numpy as np
import pandas as pd
import copy
import os

import warnings
warnings.filterwarnings("ignore")

dir_output = 'output'

# Adjust max rows to show
pd.set_option('display.max_rows', 100)

import stroke_maps
import stroke_outcome

import stroke_outcome.outcome_utilities

stroke_maps.__version__

'0.5.0'

stroke_outcome.__version__

'0.1.6'

import stroke_maps.load_data
import stroke_maps.catchment

from stroke_outcome.continuous_outcome import Continuous_outcome

Global variables to use#

limit_to_england = False

# File name for saved results
file_name = 'lsoa_base'

Load and parse data#

Load and parse hospital data#

# Get list of hopsitals and limit to those used in 2022
hospitals = stroke_maps.load_data.stroke_unit_region_lookup()

hospitals['use'] = hospitals[['use_ivt', 'use_mt']].max(axis=1)
mask = hospitals['use'] == 1
hospitals = hospitals[mask]

if limit_to_england:
    mask = hospitals['country'] == 'England'
    hospitals = hospitals[mask]

hospitals.head()

	stroke_team	short_code	ssnap_name	use_ivt	use_mt	use_msu	transfer_unit_postcode	lsoa	lsoa_code	region	region_code	region_type	country	icb	icb_code	isdn	use
postcode
SY231ER	Bronglais Hospital (Aberystwyth)	AB	Bronglais Hospital	1	0	0	nearest	Ceredigion 002A	W01000512	Hywel Dda University Health Board	W11000025	LHB	Wales	NaN	NaN	NaN	1
CB20QQ	Addenbrooke's Hospital, Cambridge	AD	Addenbrooke's Hospital	1	1	1	nearest	Cambridge 013D	E01017995	NHS Cambridgeshire and Peterborough ICB - 06H	E38000260	SICBL	England	NHS Cambridgeshire and Peterborough Integrated...	E54000056	East of England (South)	1
L97AL	University Hospital Aintree, Liverpool	AI	University Hospital Aintree	1	1	1	nearest	Liverpool 005A	E01006654	NHS Cheshire and Merseyside ICB - 99A	E38000101	SICBL	England	NHS Cheshire and Merseyside Integrated Care Board	E54000008	Cheshire and Merseyside	1
CH495PE	Arrowe Park Hospital, Wirral	AR	Wirral Arrowe Park Hospital	1	0	0	nearest	Wirral 025B	E01007251	NHS Cheshire and Merseyside ICB - 12F	E38000208	SICBL	England	NHS Cheshire and Merseyside Integrated Care Board	E54000008	Cheshire and Merseyside	1
BA13NG	Royal United Hospital Bath	BA	Royal United Hospital Bath	1	0	0	nearest	Bath and North East Somerset 008B	E01014428	NHS Bath and North East Somerset, Swindon and ...	E38000231	SICBL	England	NHS Bath and North East Somerset, Swindon and ...	E54000040	Gloucester, BSW, BNSSG and Somerset	1

Find LSOA-hospital travel time#

travel_matrix = stroke_maps.load_data.travel_time_matrix_lsoa()

# Show the first five rows and columns:
travel_matrix.iloc[:5, :5]

	B152TH	B714HJ	B95SS	BA13NG	BA214AT
LSOA
Adur 001A	173.3	179.8	171.2	161.5	152.9
Adur 001B	173.3	179.8	172.3	161.5	152.9
Adur 001C	173.3	180.9	172.3	150.8	151.9
Adur 001D	173.3	180.9	172.3	161.5	152.9
Adur 001E	174.4	180.9	173.3	150.8	151.9

# Limit LSOAs to England if required.
# Only keep LSOAs that are matched to the SICBL region type.
# SICBL are for England and LHB are for Wales.
if limit_to_england:
    # Load the LSOA-region lookup data:
    df_lsoa_regions = stroke_maps.load_data.lsoa_region_lookup()
    df_regions = stroke_maps.load_data.region_lookup()
    
    # Merge the ISDN names into the LSOA-region lookup:
    df_lsoa_regions = pd.merge(
        df_lsoa_regions.reset_index(),
        df_regions.reset_index()[['region_code', 'isdn']],
        on='region_code', how='left'
    )
    # Pick out the names of LSOA contained within SICBL:
    england_lsoa = df_lsoa_regions.loc[df_lsoa_regions['region_type'] == 'SICBL', 'lsoa']
    # Limit the travel matrix to only those LSOA:
    travel_matrix_mask = travel_matrix.index.isin(england_lsoa)
    travel_matrix = travel_matrix.loc[travel_matrix_mask].copy()

# Identify closest IVT unit
ivt_hospitals = list(hospitals[hospitals['use_ivt'] == 1].index)
df_catchment_ivt = stroke_maps.catchment.find_nearest_unit(travel_matrix[ivt_hospitals])

df_catchment_ivt.head(3)

	unit_travel_time	unit_postcode
LSOA
Adur 001A	17.6	BN25BE
Adur 001B	18.7	BN25BE
Adur 001C	17.6	BN112DH

# Identify closest MT unit
mt_hospitals = list(hospitals[hospitals['use_mt'] == 1].index)
df_catchment_mt = stroke_maps.catchment.find_nearest_unit(travel_matrix[mt_hospitals])

df_catchment_mt.head(3)

	unit_travel_time	unit_postcode
LSOA
Adur 001A	17.6	BN25BE
Adur 001B	18.7	BN25BE
Adur 001C	19.8	BN25BE

Find inter-hospital travel time#

df_transfer = stroke_maps.catchment.calculate_transfer_units(hospitals)

df_transfer.head(3)

	transfer_unit_travel_time	transfer_unit_postcode
postcode
SY231ER	135.8	CF144XW
CB20QQ	0.0	CB20QQ
L97AL	0.0	L97AL

Gather travel time information#

# Copy over IVT unit info:
df_pathway = df_catchment_ivt.copy()
df_pathway = df_pathway.rename(columns={
    'unit_travel_time': 'closest_ivt_time',
    'unit_postcode': 'closest_ivt_unit'
})

# Merge in MT unit info:
df_pathway = pd.merge(
    df_pathway, df_catchment_mt,
    left_index=True, right_index=True, how='left'
)
df_pathway = df_pathway.rename(columns={
    'unit_travel_time': 'closest_mt_time',
    'unit_postcode': 'closest_mt_unit'
})

# Merge in transfer unit info:
df_pathway = pd.merge(
    df_pathway, df_transfer,
    left_on='closest_ivt_unit', right_index=True, how='left'
)
df_pathway = df_pathway.rename(columns={
    'transfer_unit_travel_time': 'transfer_mt_time',
    'transfer_unit_postcode': 'transfer_mt_unit'
})

# Is a transfer required?
df_pathway['mt_transfer_required'] = (
    df_pathway['closest_ivt_unit'] != df_pathway['closest_mt_unit'])

Add on stroke pathway timings#

onset_to_ambulance_arrival = 60
arrival_to_ivt = 30
arrival_to_mt = 90
net_operational_delay_to_mt_for_transfer = 60

# Drip-and-ship:
df_pathway['ivt_drip_ship'] = (
    onset_to_ambulance_arrival +
    df_pathway['closest_ivt_time'] +
    arrival_to_ivt
)
df_pathway['mt_drip_ship'] = (   
    onset_to_ambulance_arrival +
    df_pathway['closest_ivt_time'] + 
    # The times in the following line are only added on if a transfer is required:
    ((net_operational_delay_to_mt_for_transfer + df_pathway['transfer_mt_time']) * df_pathway['mt_transfer_required']) +
    arrival_to_mt
)

# Mothership:
df_pathway['ivt_mothership'] = (
    onset_to_ambulance_arrival +
    df_pathway['closest_mt_time'] +
    arrival_to_ivt
)
df_pathway['mt_mothership'] = (
    onset_to_ambulance_arrival +
    df_pathway['closest_mt_time'] +
    arrival_to_mt
)

# Show df_pathway
df_pathway.head().T

LSOA	Adur 001A	Adur 001B	Adur 001C	Adur 001D	Adur 001E
closest_ivt_time	17.6	18.7	17.6	17.6	16.5
closest_ivt_unit	BN25BE	BN25BE	BN112DH	BN112DH	BN112DH
closest_mt_time	17.6	18.7	19.8	19.8	19.8
closest_mt_unit	BN25BE	BN25BE	BN25BE	BN25BE	BN25BE
transfer_mt_time	0.0	0.0	31.6	31.6	31.6
transfer_mt_unit	BN25BE	BN25BE	BN25BE	BN25BE	BN25BE
mt_transfer_required	False	False	True	True	True
ivt_drip_ship	107.6	108.7	107.6	107.6	106.5
mt_drip_ship	167.6	168.7	259.2	259.2	258.1
ivt_mothership	107.6	108.7	109.8	109.8	109.8
mt_mothership	167.6	168.7	169.8	169.8	169.8