I think I managed to solve it, but I think there is a more elegant solution out there

defmodule Scratch do
  NimbleCSV.define(MyCSVParser, [])

  # core_cols = %{
  #   date_of_birth: 'DOB',
  #   external_id: 'Plan_Member_ID',
  #   first_name: 'First_Name',
  #   last_name: 'Last_Name',
  #   phone: 'Phone'
  # }
  #
  # custom_col_pos = %{
  #   0 => 'Plan_Member_ID',
  #   1 => 'First_Name',
  #   2 => 'Phone',
  #   3 => 'Last_Name',
  #   4 => 'DOB',
  #   5 => 'HbA1c',
  #   6 => 'Hypertension',
  #   7 => 'Children',
  #   8 => 'Gender', 
  #   9 => 'Pain'
  # }
  def process_csv(file_path, %{} = custom_col_pos, %{} = core_cols) do
    file_path
    |> File.stream!(read_ahead: 1000)
    |> MyCSVParser.parse_stream([{:skip_headers, true}])
    |> Stream.map(fn line ->
      mapped_row =
        line
        |> Enum.with_index()
        |> Enum.reduce(
          %{},
          fn {cell_data, index}, acc ->
            header = Map.fetch!(custom_col_pos, index)
            column_name = if core_cols[header], do: core_cols[header], else: header

            Map.put(acc, column_name, format_col(cell_data))
          end
        )

      IO.inspect(mapped_row, label: "MappedRow")
      # {:ok, date_of_birth} = DateTimeParser.parse_date(dob)
    end)
    |> Stream.run()
  end

  defp format_col(str) do
    str |> :binary.copy() |> Macro.underscore() |> String.downcase()
  end
end

the output is (trimmed)

MappedRow: %{
  :date_of_birth => "1/1/1974",
  :external_id => "120511",
  :first_name => "jane",
  :last_name => "doe",
  :phone => "1112223333",
  'Children' => "n",
  'Gender' => "f",
  'HbA1c' => "6/3",
  'Hypertension' => "y",
  'Pain' => "y"
}

As you can see in the output I need to do some data type specific transformation i.e. HbA1c should be 6.3 not 6/3. So I guess I can use pattern matching on format_col function variants using regular expressions, but not sure if this is the best approach?