Skip to content

Unified Smoking

The smoking feature process all raw smoking information signal into unified features.

Input signals

  • Smoking_Status - categorical Smoking Status: Current, Never, Former, Unknown, Never_or_Former
  • Smoking_Intensity - How many cigarettes per day patient reports smoking in reported date. (optional signal, will use it only if exists)
  • Smoking_Duration - How many years the patient reported smoking in the reported date. (optional signal, will use it only if exists)
  • Pack_Years - Calculation of cigarettes packs per day multiply by year (Smoking_Duration * Smoking_Intensity/20). (optional signal, will use it only if exists)
  • Smoking_Quit_Date - If patient stopped smoking, the date reported stopped smoking. (optional signal, will use it only if exists)

All input signals has 1 time channel of reported date and value channel with the value.

Method

  1. The Smoking feature generator creates a patient timeline of smoking information based on all available input signals. For example, no Smoking_Status was provided in certain date but Smoking_Intensity was reported so the patient is marked as Current_Smoker within that time.
    • The smoking generator breaks the timeline based on the input signals. For example. Patient inputs were: 
      1
2
3
4
      20180101 - Reported Current Smoker
20210101 - Report Former smoker with Quit date of 20200101
20210301 - Reported Current smoker again
20250101 - Reported Former Smoker without quit date
      The feature generator will create a time line with: 
      1
2
3
4
      Assuming age 20 => 20200101 : Was Smoker
20200101 => 20200701 (middle between 20200101 to 20210301): Was Former smoker
20210101 => 20230201 (middle between 20210301 to 20250101) : Was Smoker
20230201 => today: Former Smoker
      In a case we don't have any reported smoking date, first smoking indication is "Former smoker" it uses a simple linear model (based on age reported former smoker) to estimate "exact" quit date.

Available features (Outputs)

  • One-hot smoking status: Current_Smoker, Ex_Smoker, Never_Smoker, Unknown_Smoker.
  • Smoking_Intensity: The average (median) daily smoking amount reported up to the time of prediction.
  • Pack_years: A measure of cumulative tobacco exposure, calculated either from reported pack-years or by multiplying Smoking Years by the Smoking Intensity.
  • Smok_Days_Since_Quitting: measures the number of days that have passed since the patient stopped smoking.
    • Current Smokers are assigned 0 days.
    • Never Smokers are assigned patient age in days
    • The exact quit date is given in the input signal Smoking_Quit_Date
    • If the exact quit date is missing, we estimate it using one of two methods:
      • Midpoint Estimate: If the patient's record shows they switched from being a "Current Smoker" to an "Ex-Smoker," we assume they quit exactly halfway between those two reporting dates.
      • Model Estimate: If there's no clear reporting window to use, a linear model (trained on typical quit ages) is applied to estimate the most likely quit date based on the age when the patient reported quitting.
  • Smoking_Years: The estimated total duration the patient has smoked, derived from a complete timeline of their smoking statuses (including the quit date calculation).

Please specify the requested feature output in the field smoking_features