Automated healthcare staffing
Improved methods are provided for predicting health system census over months or years and for incorporating such predictions into practitioner work schedules and predicted future practitioner needs. These improved methods permit users to interact with the predictive methods, allowing intuition and lateral goals to be conveniently incorporated into iterative updates of the predictions. These improved methods allow ‘big data’ predictive methods to be combined with human intuition by presenting users with human-interpretable adjustments to the predictor. The effects of user-selected modifications to underlying predictions and/or assumptions can be injected into the improved predictors and the effects of the modification presented back to the user, allowing the user to see the effects of various human-selected changes to hiring, admission, or other healthcare practices on long-term patient care.
1 . A method, comprising:
obtaining census data indicative of patient census across a first time period at a first temporal resolution;
generating a regression model to predict the census data as a function of time at the first temporal resolution;
applying the regression model to generate a first predicted dataset that predicts the patient census across the first time period at the first temporal resolution;
applying a trained layered artificial neural network (LANN) and a trained long short term memory artificial neural network (LSTM) to predict, at the first temporal resolution, forward-looking periods of the census data based on corresponding backward-looking periods of the census data;
determining weights to linearly combine outputs of the regression model, the LANN, and the LSTM to predict the census data;
applying the regression model to generate a second predicted dataset that predicts, at the first temporal resolution, the patient census across a second time period that is subsequent to the first time period;
applying the LANN and the LSTM to generate, based on the second predicted dataset, respective third and fourth predicted datasets that predict, at the first temporal resolution, the patient census across the second time period;
linearly combining the second, third, and fourth predicted datasets according to the weights to generate a fifth predicted dataset that predicts, at the first temporal resolution, the patient census across the second time period;
determining, based on the fifth predicted dataset, that at least one additional healthcare provider is likely to be required for a third time period within the second time period;
receiving, from a remote system, a request for the availability of a work shift during a request period that includes the third time period; and
based on determining that at least one additional healthcare provider is likely to be required for a third time period and responsive to receiving the request, sending to the remote system an indication that a work shift is available during the third time period.
2 . The method of claim 1 , further comprising:
subsequent to sending the indication that a work shift is available during the third time period to the remote system, obtaining updated census data;
updating the regression model, LANN, LSTM, and weights based on the updated census data;
applying the updated regression model, updated LANN, updated LSTM, and updated weights to generate a sixth predicted dataset that predicts, at the first temporal resolution, the patient census across the second time period;
determining, based on the sixth predicted dataset, that at least one additional healthcare provider is unlikely to be required for the third time period within the second time period;
receiving, from the remote system, an additional request for the availability of a work shift during an additional request period that includes the third time period; and
based on determining that at least one additional healthcare provider is unlikely to be required for the third time period and responsive to receiving the additional request, sending to the remote system an indication that a work shift is not available during the third time period.
3 . The method of claim 1 , wherein the regression model includes (i) at least one term related to an overall trend of the patient census over time, and (ii) one or more terms related to at least one of: day of the week, day of the year, season, month, holiday status, holiday identity, hour of the day, or shift of the day.
4 . The method of claim 1 , wherein determining, based on the fifth predicted dataset, that at least one additional healthcare provider is likely to be required for the third time period within the second time period comprises:
applying a stochastic predictor to the fifth predicted dataset to generate a plurality of simulated patient censuses; and
based on the plurality of simulated patient censuses, determining a schedule of the number of healthcare providers likely to be required across the second time period.
5 . The method of claim 4 , wherein applying the stochastic predictor to the fifth predicted dataset to generate a plurality of simulated patient censuses comprises:
determining, based on the fifth predicted dataset, a trend in the patient census across at least the second time period;
determining, across at least the second time period, one or more predicted patient admission rates into at least one department of a healthcare facility based on at least the determined trend in the patient census across at least the second time period; and
based on at least the one or more predicted patient admission rates, generating a plurality of simulated patient censuses by simulating the entry and exit of patients into and out of the at least one department of the healthcare facility.
6 . The method of claim 4 , wherein determining, based on the plurality of simulated patient censuses, the schedule of the number of healthcare providers likely to be required across the second time period comprises reducing an expected level of a cost function across the plurality of patient censuses.
7 . The method of claim 4 , wherein determining, based on the plurality of simulated patient censuses, the schedule of the number of healthcare providers likely to be required across the second time period comprises determining the schedule of the number of healthcare providers likely to be required across the second time period based on data about at least one of an anticipated amount or pattern of paid time off available during the second time period, an anticipated amount or pattern of healthcare worker vacancies during the second time period, or an anticipated amount or pattern of healthcare worker medical leave during the second time period.
8 . The method of any of claim 4 , wherein determining, based on the plurality of simulated patient censuses, the schedule of the number of healthcare providers likely to be required across the second time period comprises determining a schedule of core healthcare staff at a resolution greater than twice per day and determining a schedule of float healthcare staff at a resolution greater than twice per month.
9 . The method of claim 1 , wherein the trained LANN comprises a convolutional neural network, and wherein applying the trained LANN and trained LSTM to predict, at the first temporal resolution, forward-looking periods of the census data comprises applying (i) the corresponding backward-looking periods of the census data and (ii) the first predicted dataset to the trained LANN and to the trained LSTM.
10 . The method of claim 1 , wherein the trained LANN comprises a recurrent neural network, wherein the method further comprises:
applying a short fast Fourier transform (SFFT) to predict, at the first temporal resolution, forward-looking periods of the census data based on corresponding backward-looking periods of the census data, wherein determining weights to linearly combine outputs of the regression model, the LANN, and the LSTM to predict the census data comprises determining weights to linearly combine outputs of the regression model, the LANN, the LSTM, and the SFFT to predict the census data;
applying the SFFT to generate, based on the second predicted dataset, a seventh predicted dataset that predicts, at the first temporal resolution, the patient census across the second time period, wherein linearly combining the second, third, and fourth predicted datasets according to the weights to generate a fifth predicted dataset comprises linearly combining the second, third, fourth, and seventh predicted datasets according to the weights to generate the fifth predicted dataset.
11 . A method, comprising:
obtaining census data indicative of patient census across a first time period at a first temporal resolution;
generating a regression model to predict the census data as a function of time at the first temporal resolution;
applying the regression model to generate a first predicted dataset that predicts the patient census across the first time period at the first temporal resolution;
applying a trained layered artificial neural network (LANN) and a trained long short term memory artificial neural network (LSTM) to predict, at the first temporal resolution, forward-looking periods of the census data based on corresponding backward-looking periods of the census data;
determining weights to linearly combine outputs of the regression model, the LANN, and the LSTM to predict the census data;
applying the regression model to generate a second predicted dataset that predicts, at the first temporal resolution, the patient census across a second time period that is subsequent to the first time period;
applying the LANN and the LSTM to generate, based on the second predicted dataset, respective third and fourth predicted datasets that predict, at the first temporal resolution, the patient census across the second time period;
linearly combining the second, third, and fourth predicted datasets accordingly to the weights to generate a fifth predicted dataset that predicts, at the first temporal resolution, the patient census across the second time period;
determining, based on the fifth predicted dataset, a work schedule for a healthcare provider;
providing, via a user interface, an indication of the work schedule;
receiving, via the user interface, an indication of an update to the work schedule; and
receiving, via the user interface, an indication of an approval of the work schedule as updated according to the indicated update.
12 . The method of claim 11 , further comprising:
receiving, from a remote system subsequent to receiving the indication of the approval of the work schedule as updated according to the indicated update, a request for a work schedule for the healthcare worker; and
responsive to receiving the request, sending to the remote system an indication of the work schedule as updated according to the indicated update.
13 . The method of any of claim 11 , wherein determining, based on the fifth predicted dataset, the work schedule for the healthcare provider comprises:
determining, based on the fifth predicted dataset, a trend in the patient census across at least the second time period;
determining, across at least the second time period, one or more predicted patient admission rates into at least one department of a healthcare facility based on at least the determined trend in the patient census across at least the second time period;
based on at least the one or more predicted patient admission rates, generating a plurality of simulated patient censuses by simulating the entry and exit of patients into and out of the at least one department of the healthcare facility; and
based on the plurality of simulated patient censuses, determining a schedule of the number of healthcare providers likely to be required across the second time period.
14 . The method of claim 11 , wherein the trained LANN comprises a recurrent neural network, wherein the method further comprises:
applying a short fast Fourier transform (SFFT) to predict, at the first temporal resolution, forward-looking periods of the census data based on corresponding backward-looking periods of the census data, wherein determining weights to linearly combine outputs of the regression model, the LANN, and the LSTM to predict the census data comprises determining weights to linearly combine outputs of the regression model, the LANN, the LSTM, and the SFFT to predict the census data;
applying the SFFT to generate, based on the second predicted dataset, a seventh predicted dataset that predicts, at the first temporal resolution, the patient census across the second time period, wherein linearly combining the second, third, and fourth predicted datasets according to the weights to generate a fifth predicted dataset comprises linearly combining the second, third, fourth, and seventh predicted datasets according to the weights to generate the fifth predicted dataset.
15 . The method of claim 1 ,
wherein applying the regression model to generate a second predicted dataset comprises applying the regression model to generate a sixth predicted dataset that predicts, at the first temporal resolution, the patient census across a fourth time period that is a subset of the second time period and a seventh predicted dataset that predicts, at the first temporal resolution, the patient census across a fifth time period that is a subset of the second time period and that is subsequent to the fifth time period,
wherein applying the LANN and the LSTM to generate, based on the second predicted dataset, respective third and fourth predicted datasets comprises applying the LANN and the LSTM to generate, based on the sixth predicted dataset, respective eighth and ninth predicted datasets that predict, at the first temporal resolution, the patient census across the fifth time period, and
wherein linearly combining the second, third, and fourth predicted datasets according to the weights to generate a fifth predicted dataset comprises linearly combining the seventh, eighth, and ninth predicted datasets according to the weights to generate a tenth predicted dataset that predicts, at the first temporal resolution, the patient census across the fifth time period.
16 . The method of claim 15 , wherein applying the LANN and the LSTM to generate, based on the sixth predicted dataset, respective eighth and ninth predicted datasets comprises applying the LANN and the LSTM to generate, based on the sixth and seventh predicted datasets, the respective eighth and ninth predicted datasets, and wherein applying the trained LANN and trained LSTM to predict, at the first temporal resolution, forward-looking periods of the census data comprises applying (i) the corresponding backward-looking periods of the census data and (ii) the first predicted dataset to the trained LANN and to the trained LSTM.
17 . The method of claim 15 , wherein the trained LANN comprises a recurrent neural network, and wherein the method further comprises:
applying a short fast Fourier transform (SFFT) to predict, at the first temporal resolution, forward-looking periods of the census data based on corresponding backward-looking periods of the census data, wherein determining weights to linearly combine outputs of the regression model, the LANN, and the LSTM to predict the census data comprises determining weights to linearly combine outputs of the regression model, the LANN, the LSTM, and the SFFT to predict the census data;
applying the SFFT to generate, based on the sixth predicted dataset, an eleventh predicted dataset that predicts, at the first temporal resolution, the patient census across the fifth time period, wherein linearly combining the second, third, and fourth predicted datasets according to the weights to generate a fifth predicted dataset comprises linearly combining the seventh, eighth, ninth, eleventh predicted datasets according to the weights to generate the tenth predicted dataset.
18 . The method of claim 15 , wherein the method further comprises:
applying the LANN and the LSTM to generate respective eleventh and twelfth predicted datasets that predict, at the first temporal resolution, the patient census across the fourth time period; and
linearly combining the sixth, eleventh, and twelfth predicted datasets according to the weights to generate a thirteenth predicted dataset that predicts, at the first temporal resolution, the patient census across the fourth time period,
wherein applying the LANN and the LSTM to generate, based on the sixth predicted dataset, respective eighth and ninth predicted datasets comprises applying the LANN and the LSTM to generate, based on the thirteenth predicted dataset, the respective eighth and ninth predicted datasets.
19 . The method of claim 11 ,
wherein applying the regression model to generate a second predicted dataset comprises applying the regression model to generate a sixth predicted dataset that predicts, at the first temporal resolution, the patient census across a fourth time period that is a subset of the second time period and a seventh predicted dataset that predicts, at the first temporal resolution, the patient census across a fifth time period that is a subset of the second time period and that is subsequent to the fifth time period,
wherein applying the LANN and the LSTM to generate, based on the second predicted dataset, respective third and fourth predicted datasets comprises applying the LANN and the LSTM to generate, based on the sixth predicted dataset, respective eighth and ninth predicted datasets that predict, at the first temporal resolution, the patient census across the fifth time period, and
wherein linearly combining the second, third, and fourth predicted datasets according to the weights to generate a fifth predicted dataset comprises linearly combining the seventh, eighth, and ninth predicted datasets according to the weights to generate a tenth predicted dataset that predicts, at the first temporal resolution, the patient census across the fifth time period.
20 . The method of claim 19 , wherein applying the LANN and the LSTM to generate, based on the sixth predicted dataset, respective eighth and ninth predicted datasets comprises applying the LANN and the LSTM to generate, based on the sixth and seventh predicted datasets, the respective eighth and ninth predicted datasets, and wherein applying the trained LANN and trained LSTM to predict, at the first temporal resolution, forward-looking periods of the census data comprises applying (i) the corresponding backward-looking periods of the census data and (ii) the first predicted dataset to the trained LANN and to the trained LSTM.
21 . The method of claim 19 , wherein the method further comprises:
applying the LANN and the LSTM to generate respective eleventh and twelfth predicted datasets that predict, at the first temporal resolution, the patient census across the fourth time period; and
linearly combining the sixth, eleventh, and twelfth predicted datasets according to the weights to generate a thirteenth predicted dataset that predicts, at the first temporal resolution, the patient census across the fourth time period,
wherein applying the LANN and the LSTM to generate, based on the sixth predicted dataset, respective eighth and ninth predicted datasets comprises applying the LANN and the LSTM to generate, based on the thirteenth predicted dataset, the respective eighth and ninth predicted datasets.