predict-mortality.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Part 2: Predicting mortality in ICUs\n",
    "### Daniel Dastoor\n",
    "### Part a:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import scipy as sp\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn.linear_model import LogisticRegression\n",
    "from sklearn.feature_extraction.text import CountVectorizer, TfidfVectorizer\n",
    "from sklearn.preprocessing import MinMaxScaler\n",
    "from sklearn import metrics\n",
    "from sklearn.metrics import roc_curve\n",
    "from nltk.tokenize import word_tokenize\n",
    "from nltk.corpus import stopwords\n",
    "import string, re, multiprocessing\n",
    "from multiprocessing import Pool"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "class Data:\n",
    "    def __init__(self, data, target):\n",
    "        self.data = data\n",
    "        self.target = target\n",
    "    \n",
    "class Dataset:\n",
    "    def __init__(self, train: Data, test: Data):\n",
    "        self.train = train\n",
    "        self.test = test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "dataset = pd.read_csv(\"~/csc2548_ml4h/adult_icu.gz\", compression='gzip')\n",
    "\n",
    "min_max_scaler = MinMaxScaler()\n",
    "dataset.iloc[:,3:] = min_max_scaler.fit_transform(dataset.iloc[:,3:].values.astype(float))\n",
    "\n",
    "trainset = dataset[dataset.train == 1].copy()\n",
    "trainset.reset_index(drop=True, inplace=True)\n",
    "trainset.drop([\"train\", \"subject_id\", \"icustay_id\"], axis=1, inplace=True)\n",
    "target = trainset.pop(\"mort_icu\")\n",
    "train = Data(trainset, target)\n",
    "\n",
    "testset = dataset[dataset.train == 0].copy()\n",
    "testset.reset_index(drop=True, inplace=True)\n",
    "testset.drop([\"train\", \"subject_id\", \"icustay_id\"], axis=1, inplace=True)\n",
    "target = testset.pop(\"mort_icu\")\n",
    "test = Data(testset, target)\n",
    "\n",
    "adult_icu = Dataset(train, test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "LogisticRegression(C=1.0, class_weight=None, dual=False, fit_intercept=True,\n",
       "          intercept_scaling=1, max_iter=2000, multi_class='warn',\n",
       "          n_jobs=None, penalty='l2', random_state=None, solver='liblinear',\n",
       "          tol=0.0001, verbose=0, warm_start=False)"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "logisticRegression = LogisticRegression(penalty='l2', solver='liblinear', max_iter=2000)\n",
    "logisticRegression.fit(adult_icu.train.data.drop([\"hadm_id\"], axis=1), adult_icu.train.target)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "'AUC Score = 0.79'"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "prediction = logisticRegression.predict_proba(adult_icu.test.data.drop([\"hadm_id\"], axis=1))\n",
    "fpr, tpr, thresholds = roc_curve(adult_icu.test.target.values, prediction[:, 1])\n",
    "auc = metrics.auc(fpr, tpr)\n",
    "\n",
    "plt.figure()\n",
    "plt.plot(fpr, tpr, label='ROC curve (area = %0.2f)' % auc)\n",
    "plt.plot([0, 1], [0, 1], 'k--')\n",
    "plt.xlim([0.0, 1.0])\n",
    "plt.ylim([0.0, 1.05])\n",
    "plt.xlabel('False Positive Rate')\n",
    "plt.ylabel('True Positive Rate')\n",
    "plt.title('Receiver Operating Characteristic Curve')\n",
    "plt.legend(loc=\"lower right\")\n",
    "plt.show()\n",
    "display(\"AUC Score = \" + str(round(auc, 2)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'Accuracy = 90.84%'"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "accuracy = logisticRegression.score(adult_icu.test.data.drop([\"hadm_id\"], axis=1), adult_icu.test.target) * 100\n",
    "\"Accuracy = \" + str(round(accuracy, 2)) + \"%\""
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Comment on model performance:\n",
    "Initially, one could come to the quick conclusion that this model's performance is good, due to the \"accuracy\" of 90.84%. However, there is a large amount of class imbalance in the dataset, and as such, it is more usefull to look at the AUC score for the model. From the model's AUC score of 0.79, one can get a better sense of how the model is not performing extraordinarily well. However, with futher tuning and optimization, there is some potential in the usefulness of logistic regression on this type of normalized data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'Top 5 risk factors of mortality:'"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "['bilirubin', 'lactate', 'tempc_max', 'meanbp_mean', 'bun']"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "'Lowest 5 risk factors of mortality:'"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "['albumin', 'sysbp_min', 'platelet', 'tempc_mean', 'tempc_min']"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "feature_coefs = pd.Series(logisticRegression.coef_[0], index=adult_icu.train.data.drop([\"hadm_id\"], axis=1).columns)\n",
    "feature_coefs.sort_values(ascending=False, inplace=True)\n",
    "display(\"Top 5 risk factors of mortality:\")\n",
    "display(feature_coefs.head(5).index.tolist())\n",
    "display(\"Lowest 5 risk factors of mortality:\")\n",
    "display(feature_coefs.tail(5).index.tolist())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Top 5 risk factors of mortality:\n",
    "\n",
    "bilirubin: Forms after red blood cells break down, travelling through the liver, gallbladder, and digestive tract before excretion. Levels of high bilirubin, hyperbilirubinemia, can be indicative of a host of issues, many of which are serious.\n",
    "\n",
    "lactate: Lactate or lactic acid levels are elevated in conditions of strenuous exercise, or life-threatening conditions including heart failure, sepsis, and shock. High levels of lactate can further indicate a lack of oxygen in the blood (hypoxia).\n",
    "\n",
    "tempc_max: Maximum temprature of patient. A high maximum temprature can indicate serious issues.\n",
    "\n",
    "meanbp_mean: Average heart rate of patient. A high average heart rate can indicate serious issues.\n",
    "\n",
    "bun: Blood urea nitrogen measures the amount of urea nitrogen found in blood. This is an indication of kidney health. A high bun can be indicative of kidney or heart failure.\n",
    "\n",
    "#### Lowest 5 risk factors of mortality:\n",
    "\n",
    "albumin: Albumin is an abundant protein. High levels of albumin are caused by dehydration.\n",
    "\n",
    "sysbp_min: Minimum systolic blood pressure. Mean and max bp are much more indicative of issues.\n",
    "\n",
    "platelet: Platelets are tiny fragments of cells essential for normal blood clotting. Extreme variations in amount of platelet are not very common.\n",
    "\n",
    "tempc_mean: Average temprature of patient. Max temprate of pation is much more indicative of issues.\n",
    "\n",
    "tempc_min: Minimum temprature of patient. Max temprate of pation is much more indicative of issues."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Part b:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "dataset = pd.read_csv(\"~/csc2548_ml4h/adult_notes.gz\", compression='gzip')\n",
    "dataset = dataset[~pd.isnull(dataset.chartext)]\n",
    "\n",
    "trainset = dataset[dataset.train == 1].copy()\n",
    "trainset.reset_index(drop=True, inplace=True)\n",
    "trainset.drop([\"train\", \"subject_id\", \"icustay_id\"], axis=1, inplace=True)\n",
    "target = trainset.pop(\"mort_icu\")\n",
    "train = Data(trainset, target)\n",
    "\n",
    "testset = dataset[dataset.train == 0].copy()\n",
    "testset.reset_index(drop=True, inplace=True)\n",
    "testset.drop([\"train\", \"subject_id\", \"icustay_id\"], axis=1, inplace=True)\n",
    "target = testset.pop(\"mort_icu\")\n",
    "test = Data(testset, target)\n",
    "\n",
    "adult_notes = Dataset(train, test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "TfidfVectorizer(analyzer='word', binary=False, decode_error='strict',\n",
       "        dtype=<class 'numpy.float64'>, encoding='utf-8', input='content',\n",
       "        lowercase=True, max_df=1.0, max_features=None, min_df=1,\n",
       "        ngram_range=(1, 1), norm='l2', preprocessor=None, smooth_idf=True,\n",
       "        stop_words=None, strip_accents=None, sublinear_tf=False,\n",
       "        token_pattern='(?u)\\\\b\\\\w\\\\w+\\\\b',\n",
       "        tokenizer=<function tokenize at 0x7f5bcd4de6a8>, use_idf=True,\n",
       "        vocabulary=None)"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "stopwordList = stopwords.words(\"english\")\n",
    "\n",
    "def tokenize(s):\n",
    "    return [j for j in [re.sub(\"\\W+\", '', i) for i in word_tokenize(s.lower())] \n",
    "            if j not in stopwordList and j != '']\n",
    "\n",
    "vectorizer = TfidfVectorizer(tokenizer=tokenize)\n",
    "vectorizer.fit(adult_notes.train.data.chartext)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "num_cores = multiprocessing.cpu_count()\n",
    "num_partitions = num_cores - 1 if num_cores > 1 else 1\n",
    "\n",
    "# parallelize_dataframe adapted from \n",
    "# https://stackoverflow.com/questions/28396957/sklearn-tfidf-vectorizer-to-run-as-parallel-jobs\n",
    "def parallelize_dataframe(df, func):\n",
    "    a = np.array_split(df, num_partitions)\n",
    "    pool = Pool(num_cores)\n",
    "    df = sp.sparse.vstack(pool.map(func, a), format='csr')\n",
    "    pool.close()\n",
    "    pool.join()\n",
    "    return df\n",
    "\n",
    "train_transformed = parallelize_dataframe(adult_notes.train.data.chartext, vectorizer.transform)\n",
    "test_transformed = parallelize_dataframe(adult_notes.test.data.chartext, vectorizer.transform)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "LogisticRegression(C=1.0, class_weight=None, dual=False, fit_intercept=True,\n",
       "          intercept_scaling=1, max_iter=2000, multi_class='warn',\n",
       "          n_jobs=None, penalty='l1', random_state=None, solver='liblinear',\n",
       "          tol=0.0001, verbose=0, warm_start=False)"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "logisticRegression_b = LogisticRegression(penalty='l1', solver='liblinear', max_iter=2000)\n",
    "logisticRegression_b.fit(train_transformed, adult_notes.train.target)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "'AUC Score = 0.85'"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "prediction_b = logisticRegression_b.predict_proba(test_transformed)\n",
    "fpr_b, tpr_b, thresholds_b = roc_curve(adult_notes.test.target.values, prediction_b[:, 1])\n",
    "auc_b = metrics.auc(fpr_b, tpr_b)\n",
    "\n",
    "plt.figure()\n",
    "plt.plot(fpr_b, tpr_b, label='ROC curve (area = %0.2f)' % auc_b)\n",
    "plt.plot([0, 1], [0, 1], 'k--')\n",
    "plt.xlim([0.0, 1.0])\n",
    "plt.ylim([0.0, 1.05])\n",
    "plt.xlabel('False Positive Rate')\n",
    "plt.ylabel('True Positive Rate')\n",
    "plt.title('Receiver Operating Characteristic Curve')\n",
    "plt.legend(loc=\"lower right\")\n",
    "plt.show()\n",
    "display(\"AUC Score = \" + str(round(auc_b, 2)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'Accuracy = 91.48%'"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "accuracy_b= logisticRegression_b.score(test_transformed, adult_notes.test.target.values) * 100\n",
    "\"Accuracy = \" + str(round(accuracy_b, 2)) + \"%\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'Top 5 risk factors of mortality:'"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "['prognosis', 'cmo', 'corneal', 'dnr', 'family']"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "'Lowest 5 risk factors of mortality:'"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "['intact', 'mae', 'clear', 'diet', 'extubation']"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "feature_coefs_b = pd.Series(logisticRegression_b.coef_[0], index=vectorizer.get_feature_names())\n",
    "feature_coefs_b.sort_values(ascending=False, inplace=True)\n",
    "display(\"Top 5 risk factors of mortality:\")\n",
    "display(feature_coefs_b.head(5).index.tolist())\n",
    "display(\"Lowest 5 risk factors of mortality:\")\n",
    "display(feature_coefs_b.tail(5).index.tolist())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Comment on model performance:\n",
    "The risk factors found do make sense, especially the top 5 risk factors of mortality.\n",
    "The AUC of this model (0.85) is much more promising than that of the logicstic regression model run on normalized data. This could be as the notes for the patients captured a high amount of information and variability."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Part c:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "metadata": {},
   "outputs": [],
   "source": [
    "combined_trainset = pd.merge(adult_icu.train.data.join(adult_icu.train.target), adult_notes.train.data, on=\"hadm_id\")\n",
    "combined_target = combined_trainset.pop(\"mort_icu\")\n",
    "combined_train = Data(combined_trainset, combined_target)\n",
    "\n",
    "combined_testset = pd.merge(adult_icu.test.data.join(adult_icu.test.target), adult_notes.test.data, on=\"hadm_id\")\n",
    "combined_target = combined_testset.pop(\"mort_icu\")\n",
    "combined_test = Data(combined_testset, combined_target)\n",
    "\n",
    "combined = Dataset(combined_train, combined_test)\n",
    "\n",
    "combined_train_transformed = parallelize_dataframe(combined.train.data.chartext, vectorizer.transform)\n",
    "combined_test_transformed = parallelize_dataframe(combined.test.data.chartext, vectorizer.transform)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 116,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "LogisticRegression(C=1.0, class_weight=None, dual=False, fit_intercept=True,\n",
       "          intercept_scaling=1, max_iter=2000, multi_class='warn',\n",
       "          n_jobs=None, penalty='l1', random_state=None, solver='liblinear',\n",
       "          tol=0.0001, verbose=0, warm_start=False)"
      ]
     },
     "execution_count": 116,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "combined.train.data.drop([\"hadm_id\", \"chartext\"], axis=1, inplace=True)\n",
    "combined.test.data.drop([\"hadm_id\", \"chartext\"], axis=1, inplace=True)\n",
    "\n",
    "combined_sparse_train = sp.sparse.hstack((combined.train.data.values, combined_train_transformed))\n",
    "combined_sparse_test = sp.sparse.hstack((combined.test.data.values, combined_test_transformed))\n",
    "\n",
    "logisticRegression_c = LogisticRegression(penalty='l1', solver='liblinear', max_iter=2000)\n",
    "logisticRegression_c.fit(combined_sparse_train, combined.train.target)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 121,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "'AUC Score = 0.88'"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "prediction_c = logisticRegression_c.predict_proba(combined_sparse_test)\n",
    "fpr_c, tpr_c, thresholds_c = roc_curve(combined.test.target.values, prediction_c[:, 1])\n",
    "auc_c = metrics.auc(fpr_c, tpr_c)\n",
    "\n",
    "plt.figure()\n",
    "plt.plot(fpr_c, tpr_c, label='ROC curve (area = %0.2f)' % auc_c)\n",
    "plt.plot([0, 1], [0, 1], 'k--')\n",
    "plt.xlim([0.0, 1.0])\n",
    "plt.ylim([0.0, 1.05])\n",
    "plt.xlabel('False Positive Rate')\n",
    "plt.ylabel('True Positive Rate')\n",
    "plt.title('Receiver Operating Characteristic Curve')\n",
    "plt.legend(loc=\"lower right\")\n",
    "plt.show()\n",
    "display(\"AUC Score = \" + str(round(auc_c, 2)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 122,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'Accuracy = 91.94%'"
      ]
     },
     "execution_count": 122,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "accuracy_c = logisticRegression_c.score(combined_sparse_test, combined.test.target.values) * 100\n",
    "\"Accuracy = \" + str(round(accuracy_c, 2)) + \"%\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 123,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'Top 5 risk factors of mortality:'"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "['prognosis', 'corneal', 'stimuli', 'family', 'cmo']"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "'Lowest 5 risk factors of mortality:'"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "['good', 'diet', 'mae', 'clear', 'extubation']"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "feature_coefs_c = pd.Series(logisticRegression_c.coef_[0], index=(combined.train.data.columns.values.tolist()+vectorizer.get_feature_names()))\n",
    "feature_coefs_c.sort_values(ascending=False, inplace=True)\n",
    "display(\"Top 5 risk factors of mortality:\")\n",
    "display(feature_coefs_c.head(5).index.tolist())\n",
    "display(\"Lowest 5 risk factors of mortality:\")\n",
    "display(feature_coefs_c.tail(5).index.tolist())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Comment on how the two sources of data affect the model:\n",
    "\n",
    "While the clinical notes were an excellent source of data for a logictic regression model resulting in an AUC of 0.85, adding all of the numerical variables to the data did not result in much of an improvement (AUC increase of 0.03). Furthermore, one does not see any of the variables from the numerical data included in the top 5 or lowest 5 risk factors of mortality, showing that most of the decision is still influenced by the clinical notes data."
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}